Đọc bằng tiếng Việt. Đọc bằng tiếng Việt Đọc bằng tiếng Việt . Meet Bùi Như Mai, a 67-year-old Vietnamese grandmother who is also a YouTube influencer. “I didn’t think I was going to be a YouTube newscaster at the age of 67. But these days you will often see me at my daughter’s house, holding my grandchild while translating articles from Politico or The Atlantic into Vietnamese that I then broadcast on a news channel on YouTube. Every week I try to work my way through two, if not three articles, which takes one or two full days,” said Bùi standing in her kitchen, laughing to herself. “This is not how I imagined spending my retirement.” Bùi had noticed a lot of right wing misinformation being spread online by YouTubers in her community, particularly after Donald Trump was elected president in 2016. After she retired from her job as a software engineer at an aerospace company in 2018, she ramped up her efforts to fight back, broadcasting news on YouTube from her home in San Jose, California. Bùi reached out to me after reading my article about the unmet information needs of the Vietnamese community. One of her neighbors sent her the piece and Bùi recognized my name. I didn’t know her, but as it turns out, my dad and Bùi’s husband went to high school together in Vietnam, and have remained in touch. unmet information needs of the Vietnamese community Bùi invited me to meet with her and three dozen other Vietnamese people she had met from being a YouTuber. A few weeks later, I visited and shared resources on how to combat misinformation in the Vietnamese community. In turn, the group talked to me about how they’re trying to address misinformation, and how, because they don’t speak English, it’s been difficult getting anyone outside their community to care. It’s one of many reasons why Bùi decided to fight misinformation herself—by translating mainstream news, one YouTube video at a time. After interviewing Bùi multiple times in Vietnamese, I realized she was giving me her own oral history. What follows is Bùi’s first-hand account of how she often records her videos at 1 a.m., how she was reluctant to be on camera, the friends she’s gained and lost because of Donald Trump, and what motivates her to keep going. Bùi’s story, as told to The Markup, has been translated into English and edited to reflect the same voice and personality that comes through when she speaks in Vietnamese. Bùi’s story, as told to The Markup, has been translated into English and edited to reflect the same voice and personality that comes through when she speaks in Vietnamese. My days are filled with broadcasting news from mainstream U.S. media to my community. I don’t think the Vietnamese people in the U.S. get enough credible news. And I don’t know how to help them get credible news, except that I do the best I can with my videos. Too many receive misinformation and propaganda, especially about Donald Trump. misinformation and propaganda I started translating articles in late 2016, when Trump was elected to be President, but I first did so on Facebook. Back then, I was disappointed with the outcome of the elections and I especially didn’t understand how so many Vietnamese friends of mine could support someone who has lied about so many things and who has been such a bully. I had known them for so long and thought we shared the same values. It baffled me to see so much support for Trump on my Facebook timeline from Vietnamese people given how many articles showed his bad behavior in the lead up to the 2016 election. could support who has lied been such a bully But I had also realized that many of my Vietnamese friends did not receive particularly well-researched news about the President. At the time, I saw many Vietnamese YouTubers translating news from right-leaning news organizations like The Epoch Times and the National Enquirer that largely praised Trump. In those reports, Trump was lauded as a great businessman. He’s so rich, he must be accomplished, the videos said. He’s also a family man: he has a beautiful wife and good children who behave well. The picture painted of him was that of a successful man with a good family, of someone who is giỏi, which translates to “capable” or “accomplished,” as many Vietnamese describe him. giỏi But that is a completely different picture than the one I got from reading and subscribing to mainstream news organizations like The Mercury News, The Atlantic, The Wall Street Journal, The New York Times, Politico and The Washington Post. The picture I saw there was much more complicated. There were reports of him declaring bankruptcy multiple times, of his children almost being charged with fraud, and of his steady stream of lies. declaring bankruptcy multiple times of his children almost being charged with fraud his steady stream of lies It made me angry to think that so many Vietnamese people in the U.S. were receiving only one-sided information that praised Trump, and believing that was the whole story. I wanted to share the news I was reading with my peers, so I started posting translations of some of the stories I was reading on my Facebook timeline, adding the original story in English at the bottom. When I was young I would read the newspaper to my father every day because he was slowly losing his vision from glaucoma. Now, I am finding myself in a similar role: I want to bring legitimate news to my community. And once I started posting translations on Facebook, a lot of responses started coming from my Vietnamese circles. The majority of them were positive. A friend responded to one of my posts saying they learned a lot of English from me. Many people also told me that reading my translations reminded them of when they were kids in Vietnam: The language they spoke in their homes in Vietnam had been standardized by the government over the years and most Vietnamese media no longer use the same Vietnamese we learned as kids. The Vietnamese words I use convey a sense of trust and community for people whose feelings and thoughts are still deeply rooted in their past. The hard thing for the translation is not the English but the Vietnamese. You have to translate not just the words, but also translate it in cultural ways for the Vietnamese people to relate to it. It’s less about translation and more about interpretation. Sometimes it takes me 15 minutes to rewrite one sentence from English into Vietnamese to get both the gist of the article but to also make sure it sounds like something a Vietnamese person would actually write, say, and fully comprehend. For instance, I try to use common Vietnamese idioms to help better translate articles. I’ve used phrases like anh hùng xó bếp, which translates to “heroes in the kitchen” and refers to people who talk a big game but only in the safety of their kitchen. Or đầu trâu mặt ngựa, which translates to “head of a buffalo and the face of a horse” and means immoral people. anh hùng xó bếp đầu trâu mặt ngựa Of course, posting about Trump also invites a lot of criticism. I was bullied by some of my community members for my translations. Once, a woman who didn’t know me started threatening to beat me in the comments of my posts. She was insistent and would find ways to circumvent being blocked by Facebook—she would purposely misspell Vietnamese words, enough to not automatically be booted off the platform, but their meaning was still clear. My husband thinks she threatened to kill me. My daughters became very worried and one of them ended up reporting the commenter to Facebook. “I don’t have a face for TV. If I broadcast my face they will just start making fun of me” After doing these translations for four years on Facebook, I had a steady following, with some of the more popular ones garnering up to 20-30 comments. Last year I received a call from Ái Vân, a famous singer from Vietnam and a friend of mine who also lives in San Jose. We first met through a mutual friend fifteen years ago and recently became closer friends when she found out I wasn’t fond of Trump. She had recently been recruited by a YouTube show to speak about political issues in the US to a Vietnamese audience. The show is called Người Việt, which translates to Vietnamese People and is separate from a national Vietnamese language newspaper that’s existed since the 1970s. Vân had seen my Facebook posts and thought I should be on their channel, too. Ái Vân “Absolutely not!” I told her. “I don’t have a face for TV. If I broadcast my face they will just start making fun of me for being old or the way I look, not for what my translations were like.” But the founder of the channel followed up. He called once. Then another time. Then two more times. Each time I told him “no.” Then he recruited his friend, a mathematics professor at a university in Houston, Nguyễn Dinh Minh Quốc, to try to get me on board. We didn’t know each other, but he called me, still I said no. He called me again, I said no once more. When the professor called me a third time, he caught me off guard: I was taking care of my grandson who was a baby at the time and had been crying all day. I was so exhausted from trying to calm my grandson that I finally relented and said I’d give it a try. The first time I contributed to Người Việt was them interviewing me on a live stream. The recording stream has now been watched more than 10,000 times. It was a big surprise to me. The two men running the channel decided then and there that I needed my own show. With the approval of my husband and my daughters, I said yes. The trial run put them at ease and they were excited to call me a Vietnamese influencer. I now broadcast YouTube videos with news from legitimate news organizations twice a week. I’m not a journalist and I never have been but I invented a system for me to make it all work. The folks from the Người Việt channel helped me figure out what to buy—a green screen, some lights, a microphone—and I’ve set up a small studio in our Buddhist meditation room. And so I record from that room, staring at my laptop which sits on a stack of books. Every Monday and Thursday I pick stories to feature on my news videos. Usually it’s current news, like the ongoing conflict in Ukraine. Other times I pick feature articles: once I translated an article about happiness. But there’s nothing that brings in as many clicks among my Vietnamese-speaking audience as stories about Donald Trump. Most of the stories I pick about the former president are critical about him because I know my community is already getting the news that praises him from other channels and because I don’t want him to be re-elected. I get reactions from all kinds of people, many who dislike him but some who support him. I try to translate at least two stories, if not three, for each video and that can be an arduous undertaking. On weekdays, I start translating a story while I’m babysitting my infant granddaughter. Sometimes it takes me so long to translate an article because I have to go back and forth between taking care of my grand kid and doing this work. Sometimes an article that I like more publishes when I’m halfway or two thirds through a translation, forcing me to start my translation all over again. The average time it takes to record a video is about 3 hours because I have to do multiple takes. Since I do not know how to edit I have to re-record the whole video if I accidentally mess up and say the wrong word. Sometimes I end up working long after I hand my granddaughter back over to her parents and late into the night until 1 or 2 am, just to get the translation right. Even though I worked in an office setting as an engineer for close to three decades in the U.S., there are some articles that really push the limits of my English and I have to ask my daughters for help. Here is an example of a paragraph in a New York Times piece that gave me trouble: piece Credit: Bùi Như Mai And even after translating the stories, I still have a lot of work to do. I have to write my introduction for the video and my commentary on the articles. If I don’t get done until late into the night until 1 or 2 am, I usually take a shower, put on the work clothes that I used to wear before I retired, a nice pearl necklace from a collection I’ve put together over the years, and then start shooting my video from my home studio. Credit:YouTube A network of friends helping each other A lot has changed since I started broadcasting news videos on YouTube. For one, I’m no longer close to friends who are staunch Trump supporters. I know much of the American public was polarized by Trump. And my community is not any different. But I’ve also found new friends through my work. Over the past two years I’ve met many new friends who are politically active. We talk about thằng Trump—don’t respect him so we call him “boy” Trump, not ông Trump which would be the appropriate honorific. We gather in my backyard for celebrations, have potlucks with Vietnamese food and discuss how we can help our community learn more. thằng ông I also have fans now. When I recently announced that I was going on a six week vacation in Vietnam some of my viewers told me in the comments that they were a little sad. Through that announcement video I also learned that I had regular viewers in Vietnam. One of them who lives in the city of Vung Tàu wants to meet me while I’m on holiday. All of this is tiring work. But at least we do something, rather than nothing. The other day I was recognized in the parking lot of a supermarket. “Mai Bùi? Is this you?” The person asked. And my heart stopped for a few seconds. I wasn’t sure if this person was going to yell at me in anger or be kind. Trump has become such a divisive figure in our community, it can be hard to decipher whether someone sees what I do as a threat or as a service. “Yes, it’s me,” I said. “I watch you all the time! Thank you for what you do.” I felt relieved. Credits Bùi Như Mai
Lam Thuy Vo, Investigative Reporter Bùi Như Mai Lam Thuy Vo, Investigative Reporter Lam Thuy Vo Illustration Gabriel Hongsdusit Gabriel Hongsdusit Gabriel Hongsdusit Photography Lam Thuy Vo Lam Thuy Vo Lam Thuy Vo Engagement Maria Puertas Maria Puertas Maria Puertas Editing Sisi Wei
Michael Reilly Sisi Wei Sisi Wei Michael Reilly Michael Reilly Also published here here Photo by Julian Rösner on Unsplash Julian Rösner Unsplash

This Vietnamese Grandmother Is Using YouTube to Fight Misinformation

Fighting for Algorithmic Justice: The Struggle to Unmask AI

Introducing Nightshade: A Tool Helping Artists to Fight Against AI 

VCs, DIDs, and the Fight for Digital Autonomy

Fighting Piracy With SEO: How I’m Turning Malware Traffic Into Safer Gaming Choices

I’m not a developer. Not in the traditional sense, anyway. I didn’t get a fancy comp-sci degree. I didn’t intern at Google. I didn’t wake up one day thinking, “Let me master the React lifecycle or deploy a CI/CD pipeline before breakfast.” What I am is a guy who’s been broke, burned out, and trying to code his way out of a small-town cage with Wi-Fi just good enough to crash halfway through a Git pull. I taught myself the web one Stack Overflow thread at a time. I know just enough HTML to build something scrappy and just enough Python to break it. I’ve freelanced. I’ve stared at my screen until my vision blurred. am And I know this: devs need more than clean code. You need clean boundaries, too. Productivity Isn’t Your God We glorify hustle like it’s sacred. "Crush it." "Ship fast." "Outwork everyone." You know what’s sacred? Your sanity. There’s a difference between being committed and being consumed. Too many devs I’ve met are constantly tired, constantly plugged in, constantly chasing the next feature push like it’s going to fix their life. It won’t. You’ve got to build rest into your routine like you build testing into your code. Or else something breaks. And usually, it’s you. You’re Allowed to Not Know Impostor syndrome? That’s not a bug. That’s the whole damn industry. Nobody knows everything. Not your tech lead. Not that one guy on Twitter who builds 8 SaaS projects a week. Not even the guy who built the thing you’re working on now. I used to feel like a fraud every time I opened VS Code. Still do some days. But here’s the thing: asking questions doesn’t make you weak. It makes you real. If you’re stuck, ask. If you’re lost, Google it. If you can’t find the answer, build the damn answer Don’t Trade Your Life for a Launch Yeah, shipping is cool. Startups are sexy. Getting that dopamine hit from a successful deploy feels good. But too many devs are sprinting toward a finish line that moves every time they get close. You don’t have to sacrifice your friendships, your body, your late nights, or your sanity for another SaaS launch. There’s always going to be another sprint. Protect your peace. Side Projects Should Feed You, Not Drain You Side projects are supposed to be your playground, not another job. Build stuff that excites you. Make something dumb. Hack together something that doesn’t scale. If your side project starts feeling like a second boss, you’ve already lost. You don’t need to monetize every idea. Some stuff is just for joy. That’s allowed. Find People Who Get It Loneliness is real in this space. Especially if you’re freelancing or remote. Find a Discord. A subreddit. A group chat. Somewhere you can ask, “Does anyone else feel like they’re losing it trying to deploy this API?” Code can be solitary. But you don’t have to suffer alone. Tech Won’t Save You It won’t fix your heartbreak. It won’t fill your emptiness. It won’t replace the conversation you’re not having or the life you’re not living. I used to think if I could just build enough—websites, businesses, side hustles—I’d finally feel okay. That I’d finally get free. build enough But freedom isn’t found in output. It’s found in honesty. You can learn every framework on the planet and still feel stuck. So be real with yourself. What are you actually chasing? Burnout Isn’t a Badge of Honor If you’re exhausted all the time, that’s not grind culture. That’s a warning sign. Sleep. Hydrate. Walk away from the damn screen. Let your brain reset. Burnout isn’t proof you care. It’s proof you didn’t. Care for yourself like you care about your uptime. You’re Not Behind Everyone’s posting wins. New jobs. Product launches. Y Combinator rejections that still go viral. It’s easy to feel like you’re not doing enough. But listen: social media is a highlight reel. Not the whole picture. Your journey isn’t late. It’s yours. Keep building. Not because you’re trying to prove something. But because you still believe something beautiful might come out of this mess. I’m Sahadat hossen (Sagor). I write from a town most people forget. I build stuff with half-broken tools. I mess up. I restart. And I keep going. If you’re a dev and you’re tired—really tired—I see you. You’re not alone. You’re not failing. You’re just human. Build slow. Live loud. And don’t forget why you started. —Sahadat.

Code, Burnout, and Late-Night Debugging: Life Advice for Devs From Someone Still Figuring It Out

Table of Links Abstract and 1. Introduction Abstract and 1. Introduction 2. Community Challenges Overview and 2.1 CCKS 2. Community Challenges Overview and 2.1 CCKS 2.2 CHIP and 2.3 CCIR, CSMI, CCL and DCIC 2.2 CHIP and 2.3 CCIR, CSMI, CCL and DCIC 3. Evaluation Tasks Overview and 3.1 Information Extraction 3. Evaluation Tasks Overview and 3.1 Information Extraction 3.2 Text Classification and Text Similarity 3.2 Text Classification and Text Similarity 3.3 Knowledge Graph and Question Answering 3.3 Knowledge Graph and Question Answering 3.4 Text Generation and Knowledge Reasoning and 3.5 Large Language Model Evaluation 3.4 Text Generation and Knowledge Reasoning and 3.5 Large Language Model Evaluation 4. Translational Informatics in Biomedical Text Mining 4. Translational Informatics in Biomedical Text Mining 5. Discussion and Perspective 5.1. Contributions of Community Challenges 5.1. Contributions of Community Challenges 5.2. Limitations of Current Community Challenges 5.2. Limitations of Current Community Challenges 5.3. Future Perspectives in the Era of Large Language Models, and References 5.3. Future Perspectives in the Era of Large Language Models, and References Figure Legends and Tables Figure Legends and Tables Figure Legends and Tables This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. This paper is available on arxiv under CC BY-NC-ND 4.0 DEED license. available on arxiv Authors:
(1) Hui Zong, Joint Laboratory of Artificial Intelligence for Critical Care Medicine, Department of Critical Care Medicine and Institutes for Systems Genetics, Frontiers Science Center for Disease related Molecular Network, West China Hospital, Sichuan University, Chengdu, 610041, China and the author contributed equally;
(2) Rongrong Wu, Joint Laboratory of Artificial Intelligence for Critical Care Medicine, Department of Critical Care Medicine and Institutes for Systems Genetics, Frontiers Science Center for Disease related Molecular Network, West China Hospital, Sichuan University, Chengdu, 610041, China and the author contributed equally;
(3) Jiaxue Cha, Shanghai Key Laboratory of Signaling and Disease Research, Laboratory of Receptor-Based Bio-Medicine, Collaborative Innovation Center for Brain Science, School of Life Sciences and Technology, Tongji University, Shanghai, 200092, China;
(4) Erman Wu, Joint Laboratory of Artificial Intelligence for Critical Care Medicine, Department of Critical Care Medicine and Institutes for Systems Genetics, Frontiers Science Center for Disease related Molecular Network, West China Hospital, Sichuan University, Chengdu, 610041, China;
(5) Jiakun Li, Joint Laboratory of Artificial Intelligence for Critical Care Medicine, Department of Critical Care Medicine and Institutes for Systems Genetics, Frontiers Science Center for Disease related Molecular Network, West China Hospital, Sichuan University, Chengdu, 610041, China and Department of Urology, West China Hospital, Sichuan University, Chengdu, 610041, China;
(6) Liang Tao, Faculty of Business Information, Shanghai Business School, Shanghai, 201400, China;
(7) Zuofeng Li, Takeda Co. Ltd., Shanghai, 200040, China;
(8) Buzhou Tang, Department of Computer Science, Harbin Institute of Technology, Shenzhen, 518055, China;
(9) Bairong Shen, Joint Laboratory of Artificial Intelligence for Critical Care Medicine, Department of Critical Care Medicine and Institutes for Systems Genetics, Frontiers Science Center for Disease related Molecular Network, West China Hospital, Sichuan University, Chengdu, 610041, China and a Corresponding author. Authors: Authors: (1) Hui Zong, Joint Laboratory of Artificial Intelligence for Critical Care Medicine, Department of Critical Care Medicine and Institutes for Systems Genetics, Frontiers Science Center for Disease related Molecular Network, West China Hospital, Sichuan University, Chengdu, 610041, China and the author contributed equally; (2) Rongrong Wu, Joint Laboratory of Artificial Intelligence for Critical Care Medicine, Department of Critical Care Medicine and Institutes for Systems Genetics, Frontiers Science Center for Disease related Molecular Network, West China Hospital, Sichuan University, Chengdu, 610041, China and the author contributed equally; (3) Jiaxue Cha, Shanghai Key Laboratory of Signaling and Disease Research, Laboratory of Receptor-Based Bio-Medicine, Collaborative Innovation Center for Brain Science, School of Life Sciences and Technology, Tongji University, Shanghai, 200092, China; (4) Erman Wu, Joint Laboratory of Artificial Intelligence for Critical Care Medicine, Department of Critical Care Medicine and Institutes for Systems Genetics, Frontiers Science Center for Disease related Molecular Network, West China Hospital, Sichuan University, Chengdu, 610041, China; (5) Jiakun Li, Joint Laboratory of Artificial Intelligence for Critical Care Medicine, Department of Critical Care Medicine and Institutes for Systems Genetics, Frontiers Science Center for Disease related Molecular Network, West China Hospital, Sichuan University, Chengdu, 610041, China and Department of Urology, West China Hospital, Sichuan University, Chengdu, 610041, China; (6) Liang Tao, Faculty of Business Information, Shanghai Business School, Shanghai, 201400, China; (7) Zuofeng Li, Takeda Co. Ltd., Shanghai, 200040, China; (8) Buzhou Tang, Department of Computer Science, Harbin Institute of Technology, Shenzhen, 518055, China; (9) Bairong Shen, Joint Laboratory of Artificial Intelligence for Critical Care Medicine, Department of Critical Care Medicine and Institutes for Systems Genetics, Frontiers Science Center for Disease related Molecular Network, West China Hospital, Sichuan University, Chengdu, 610041, China and a Corresponding author.

Figure Legends and Tables for Our Biomedical Text Mining Research

Site Reliability Engineering (SRE) is one of the key pillars for organizations. SRE teams are responsible for maintaining the system's scalability and reliability. One of the key challenges SRE teams face is dealing with alert floods, parsing cryptic logs, and the pressure of SLA timers. These challenges make Root Cause Analysis (RCA) of an incident really tough. With the increasing complexity of distributed infrastructure, identifying RCA and resolving incidents becomes more difficult. Because conventional troubleshooting methods require manual log analysis and the review of multiple data sources, they are very time-consuming and demand a large employee workforce. In this article we will examine how Artificial Intelligence (AI) is benefiting Root Cause Analysis (RCA) in incident management by automating processes, reducing resolution time, and improving overall system reliability. This article delves into the techniques used and challenges faced. Overview For SRE teams, AI integration into RCA has emerged as a very powerful solution. RCA systems using machine learning, natural language processing, and event correlation to automate root cause identification helps the SRE team to speed up the incident resolution. SRE teams having AI driven RCA can shift from reactive to proactive approach for system reliability. This will further help them reduce the operational cost while improving service availability. The following sections detail AI techniques used for root cause analysis. AI Techniques for Root Cause Analysis Incident Detection (Monitoring and Alerting) As part of incident monitoring, you can use AI integrated systems to continuously monitor logs, metrics, and alerts from various sources to detect anomalies or unusual system behaviors. You can use real-time event streaming platforms like Apache Kafka and AWS Kinesis to aggregate logs and metrics across distributed systems. By using supervised and unsupervised learning techniques, you can train AI models on historical incident data. This will establish normal system behavior baselines and the detection of deviations. You can also use techniques such as Hidden Markov Models (HMMs) and Gaussian Mixture Models (GMMs) which can help detect temporal patterns in log data, while deep learning architectures like Transformer-based anomaly detection models enhance accuracy. API Request API Request GET /api/incident-detection GET /api/incident-detection Example Response Example Response { 

 "status":

 "success",

 "detected_incidents": [

   {"id": "INC001", "timestamp": "2024-02-28T10:15:00Z", "severity": "High"},

   {"id": "INC002", "timestamp": "2024-02-28T10:20:00Z", "severity": "Medium"}

 ]

} { 

 "status":

 "success",

 "detected_incidents": [

   {"id": "INC001", "timestamp": "2024-02-28T10:15:00Z", "severity": "High"},

   {"id": "INC002", "timestamp": "2024-02-28T10:20:00Z", "severity": "Medium"}

 ]

} Log Analysis (NLP & Error Detection) At this stage, it's important to review the logs for the incidents that were captured earlier. To extract meaningful information from the unstructured log data, techniques like natural language processing (NLP) can be used. This process is supported by widely used log analysis platforms such as Splunk, the Elastic Stack, or Datadog.  These tools often use advanced learning systems, using models like BERT, GPT-style transformers, and LSTM networks, to sort through the logs, spot patterns that suggest errors, and pull out useful information that helps figure out the underlying causes of issues.These models can parse log entries, detect anomalies through semantic analysis, and apply Named Entity Recognition (NER) to identify critical infrastructure components, error codes, and affected services. Additionally, you can use clustering algorithms like K-Means and hierarchical clustering to group similar log patterns, enhancing the accuracy of root cause analysis. API Request API Request POST /api/log-analysis POST /api/log-analysis Example Request Body Example Request Body { 

 "incident_id": "INC001",

 "log_data": "Error 503: Service Unavailable at /api/payment",

 "context": "E-commerce checkout failure"

} { 

 "incident_id": "INC001",

 "log_data": "Error 503: Service Unavailable at /api/payment",

 "context": "E-commerce checkout failure"

} Example Response Example Response { 

 "status": "success",

 "parsed_logs": ["Service Unavailable at /api/payment", "Possible cause: High API latency"],

 "severity": "High",

 "metrics": ["Response Time", "API Latency", "CPU Usage"]

} { 

 "status": "success",

 "parsed_logs": ["Service Unavailable at /api/payment", "Possible cause: High API latency"],

 "severity": "High",

 "metrics": ["Response Time", "API Latency", "CPU Usage"]

} Anomaly Detection (ML Models & Adaptive Thresholds) Anomaly detection plays a crucial role in this phase of root cause analysis. To effectively detect irregularities in distributed systems, you need to combine traditional statistical methods—like Z-score analysis or Isolation Forest—with machine learning approaches such as LSTMs and Autoencoders. Statistical analysis offers valuable insights, whether you're looking at individual variables or exploring relationships across multiple factors. When working with time-based data, deep learning models are especially useful for recognizing patterns that unfold over time. For detecting unusual behavior in time series, methods like Seasonal Hybrid ESD are effective for identifying seasonality-based anomalies, while tools such as Prophet can be used to anticipate irregular trends and forecast deviations. You can also use advanced methods like Dynamic Thresholding with moving averages and Bayesian Change Point Detection (BCPD). Integration with log analysis outputs enables the anomaly detection models to prioritize specific metrics that align with identified log patterns. For example, if log analysis indicates "High API latency," anomaly detection can apply focused monitoring on response times, CPU usage, and memory allocation. This integrated approach improves precision in root cause identification by correlating log-derived insights with detected anomalies in system metrics. API Request API Request POST /api/anomaly-detection POST /api/anomaly-detection Example Request Body Example Request Body { 

 "incident_id": "INC001",

 "related_log_analysis": "High API latency",

 "metrics": ["Response Time", "API Latency", "CPU Usage"],

 "thresholds": { "CPU Usage": "80%", "Response Time": "1s", "API Latency": "500ms"}

} { 

 "incident_id": "INC001",

 "related_log_analysis": "High API latency",

 "metrics": ["Response Time", "API Latency", "CPU Usage"],

 "thresholds": { "CPU Usage": "80%", "Response Time": "1s", "API Latency": "500ms"}

} Example Response Example Response { 

 "status": "success",

 "anomalies_detected": [

   {"metric": "CPU Usage", "value": "95%", "threshold": "80%"},

   {"metric": "Response Time", "value": "2s", "threshold": "1s"}

 ]

} { 

 "status": "success",

 "anomalies_detected": [

   {"metric": "CPU Usage", "value": "95%", "threshold": "80%"},

   {"metric": "Response Time", "value": "2s", "threshold": "1s"}

 ]

} Pattern Recognition (Decision Trees, Neural Networks) In this step, you can use Machine learning (ML) models to analyze historical incident data to identify recurring failure patterns using techniques such as Decision Trees, Random Forests, Gradient Boosting, and Deep Neural Networks. Ensemble learning methods, including Bagging and Boosting, enhance prediction accuracy by combining multiple weak models into a strong predictive model. You can use techniques like Generative Adversarial Networks (GANs) to simulate potential system failures, creating synthetic datasets that improve the robustness of AI models under rare or edge case scenarios. Use of models like XGBoost and CatBoost can offer high performance in handling structured data, enabling quick pattern recognition and anomaly prediction in large scale distributed systems. To assist in automatically selecting the best model, fine-tuning hyperparameters, and deploying models, you can also use AutoML frameworks such as Google's AutoML and Microsoft's Azure AutoML. This will allow SRE teams to integrate ML based pattern recognition into their incident management workflows. API Request API Request POST /api/pattern-recognition POST /api/pattern-recognition Example Request Body Example Request Body { 

 "incident_id": "INC001",

 "anomalies_detected": [

   {"metric": "CPU Usage", "value": "95%", "threshold": "80%"},

   {"metric": "Response Time", "value": "2s", "threshold": "1s"}

 ]

} { 

 "incident_id": "INC001",

 "anomalies_detected": [

   {"metric": "CPU Usage", "value": "95%", "threshold": "80%"},

   {"metric": "Response Time", "value": "2s", "threshold": "1s"}

 ]

} Example Response Example Response { 

 "status": "success",

 "patterns_detected": [

   "Configuration Change Failure",

   "Database Deadlocks",

   "Memory Leaks"

 ]

} { 

 "status": "success",

 "patterns_detected": [

   "Configuration Change Failure",

   "Database Deadlocks",

   "Memory Leaks"

 ]

} Event Correlation (Graph Neural Networks) In this step to establish relationships between the incidents, you need to use AI to correlate multiple sources of data such as logs, metrics, alerts. You can use graph based RCA models to utilize techniques like Graph Neural Networks (GNN), Bayesian Networks, and Markov Random Fields (MRF) to visualize complex service dependencies and model probabilistic relationships between components. You can also use Temporal Graph Convolutional Networks (T-GCN) and Dynamic Graph Attention Networks (DGAT) that enhance event correlation by capturing both time dependent relationships and dynamic changes in the infrastructure topology. These models help predict cascading failures by analyzing event sequences and identifying potential failure chains by using techniques such as Path Ranking Algorithms (PRA) and Shortest Path Algorithms (Dijkstra's, A*). API Request API Request POST /api/event-correlation POST /api/event-correlation Example Request Body Example Request Body { 

 "incident_id": "INC001",

 "patterns_detected": [

   "Configuration Change Failure",

   "Database Deadlocks",

   "Memory Leaks"

 ]

} { 

 "incident_id": "INC001",

 "patterns_detected": [

   "Configuration Change Failure",

   "Database Deadlocks",

   "Memory Leaks"

 ]

} Example Response Example Response { 

 "status": "success",

 "incident_id": "INC001",

  "related_events": [

   "Database Latency Spike",

   "API Gateway Failure",

    "Service Restart Triggered"

 ],

 "root_cause": "Database Latency Spike"

} { 

 "status": "success",

 "incident_id": "INC001",

  "related_events": [

   "Database Latency Spike",

   "API Gateway Failure",

    "Service Restart Triggered"

 ],

 "root_cause": "Database Latency Spike"

} Automated Remediation (Self Healing Actions) AI integrated RCA systems not only help you find the root cause of a problem, but also help with its resolution. AI integrated systems can take action on their own using decision rules and learning algorithms. For instance, you can configure them to restart services, undo configuration changes, or adjust system resources automatically to keep the system running smoothly. You can further use more advanced AI methods like Deep Q-Networks (DQN) and Proximal Policy Optimization (PPO).These methods help the system learn from past incidents so it can make smarter decisions in the future—balancing trying new things with sticking to what works. You can also use tools like Rundeck and Ansible that are often used to run these fixes automatically, without human input. AI driven platforms can tie together monitoring and response so the system can detect problems, take action, confirm the fix worked, and update its own settings all without needing a person to step in. These systems can also predict resource needs and adjust things like computing power or storage on the fly, making sure there’s enough to recover from issues without wasting resources. API Request API Request POST /api/incident-response POST /api/incident-response Example Request Body Example Request Body { 

 "incident_id": "INC001",

 "root_cause": "Database Latency Spike",

 "severity": "Critical",

 "auto_remediation_enabled": true

} { 

 "incident_id": "INC001",

 "root_cause": "Database Latency Spike",

 "severity": "Critical",

 "auto_remediation_enabled": true

} Example Response Example Response { 

 "status": "success",

 "remediation_action": "Scaled up database instances",

 "status": "Resolved",

 "root_cause_handled": "Database Latency Spike",

  "next_steps": [

   "Monitor database performance",

    "Optimize slow queries"

 ]

} { 

 "status": "success",

 "remediation_action": "Scaled up database instances",

 "status": "Resolved",

 "root_cause_handled": "Database Latency Spike",

  "next_steps": [

   "Monitor database performance",

    "Optimize slow queries"

 ]

} Benefits of Using AI in RCA for SRE Benefits of Using AI in RCA for SRE Integrating AI into RCA provides multiple benefits to the SRE teams. AI improves the efficiency and accuracy of incident resolution. Some of the benefits are listed below: Faster Incident Resolution – AI can analyze logs and detect irregularities in real time.This will reduce the time taken to identify and resolve issues which will further improve user experience.
Continuous Learning from Past Incidents – AI systems can continuously learn from the past incidents. Thus improving RCA leading to faster incident resolution.
Proactive Problem Detection – SRE teams can take preventive actions by using AI’s predictive analytics that can detect early warning signs of failures.
Improved Accuracy – AI integrated systems can improve accuracy by automatically detecting the root cause. This will further reduce human errors. Thus reducing false positives.
Reduced Operational Costs – AI integrated RCA systems will reduce manual investigations. This in turn improves operating efficiency and lesser operational costs.
Scalability – AI driven RCA can handle large amounts of log data and system metrics across distributed infrastructures. This will help efficient analysis at a large scale.
Better Collaboration – AI provided information provides structured data that facilitates communication between teams, ensuring more efficient incident resolution. Faster Incident Resolution – AI can analyze logs and detect irregularities in real time.This will reduce the time taken to identify and resolve issues which will further improve user experience. Faster Incident Resolution Continuous Learning from Past Incidents – AI systems can continuously learn from the past incidents. Thus improving RCA leading to faster incident resolution. Continuous Learning from Past Incidents Proactive Problem Detection – SRE teams can take preventive actions by using AI’s predictive analytics that can detect early warning signs of failures. Proactive Problem Detection Improved Accuracy – AI integrated systems can improve accuracy by automatically detecting the root cause. This will further reduce human errors. Thus reducing false positives. Improved Accuracy Reduced Operational Costs – AI integrated RCA systems will reduce manual investigations. This in turn improves operating efficiency and lesser operational costs. Reduced Operational Costs Scalability – AI driven RCA can handle large amounts of log data and system metrics across distributed infrastructures. This will help efficient analysis at a large scale. Scalability Better Collaboration – AI provided information provides structured data that facilitates communication between teams, ensuring more efficient incident resolution. Better Collaboration Challenges of AI in RCA Challenges of AI in RCA Even though there are several advantages of AI in RCA, integrating AI in RCA presents several challenges: Data Quality and Volume – AI models requires large amount of good quality data to operate effectively and provide quality recommendations.
Invisible Decision making – AI’s decision making process is not visible to the team, making it hard for the SRE team to trust the recommendations provided.
False Positives and Negatives – AI can incorrectly identify RCA. This can further delay the RCA and can impact business. This will further lead to operating losses for the organizations.
Integration Complexity – AI must integrate correctly with the existing monitoring, logging, and incident management tools. This can be challenging and expensive for the organizations.
Security and Compliance – AI systems must adhere to security and privacy regulations since they are handling sensitive logs and system data. Data Quality and Volume – AI models requires large amount of good quality data to operate effectively and provide quality recommendations. Data Quality and Volume Invisible Decision making – AI’s decision making process is not visible to the team, making it hard for the SRE team to trust the recommendations provided. Invisible Decision making False Positives and Negatives – AI can incorrectly identify RCA. This can further delay the RCA and can impact business. This will further lead to operating losses for the organizations. False Positives and Negatives Integration Complexity – AI must integrate correctly with the existing monitoring, logging, and incident management tools. This can be challenging and expensive for the organizations. Integration Complexity Security and Compliance – AI systems must adhere to security and privacy regulations since they are handling sensitive logs and system data. Security and Compliance To overcome these challenges, organizations need to invest in explainable AI (XAI). They should also need to keep a human in the loop approach for validation, and continuously refine AI models with real world data. Conclusion Conclusion AI is transforming root cause analysis in SRE by automating log analysis, event correlation, anomaly detection, and incident resolution. By using AI integrated systems, SRE teams can resolve incidents faster, reduce manual troubleshooting efforts, improve system reliability and reduce operating costs. Although there are multiple challenges with AI in RCA, advancements in AI explainability will continue to enhance RCA accuracy. As AI technology evolves, SRE teams that use AI integrated RCA will be better positioned to manage complex distributed infrastructures with efficiency and confidence.

This Is How AI Figures Out Whether Your Systems Are Crashing

Welcome to HackerNoon’s Writing Prompts! Would you like to take a stab at answering some of these questions? The link for the template is HERE. Welcome to HackerNoon’s Writing Prompts! Would you like to take a stab at answering some of these questions? The link for the template is HERE. HackerNoon The link for the template is HERE. 1. What is your company in 2–5 words? Denis Alaev: EODHD APIs is a top-rated financial data provider. Denis Alaev: Denis Alaev: Denis Alaev top-rated top-rated 2. Why is now the time for your company to exist? The demand for reliable and affordable financial data is higher than ever. As more businesses and individuals rely on data-driven decision-making, there's a growing need for accessible APIs that deliver accurate financial data without the hefty costs of traditional providers. 3. What do you love about your team, and why are you the ones to solve this problem? Our team combines deep technical expertise with product development experience from international tech companies. We understand both the technical challenges of financial data and the user experience side, which makes us uniquely positioned to build a product that's both robust and easy to use. 4. If you weren’t building your startup, what would you be doing? I’d likely still be building products — either in big tech or a fast-paced startup. With my background in computer science and product management, I'd continue solving technical problems that matter. 5. At the moment, how do you measure success? What are your metrics? At the moment, we measure success primarily by the number of active users — EODHD APIs currently has tens of thousands, which is our key product metric. EODHD APIs Additionally, we closely monitor customer satisfaction through our Trustpilot rating and user feedback. Positive reviews and high satisfaction scores are strong indicators that we’re delivering real value and meeting user expectations. 6. In a few sentences, what do you offer to whom? We provide high-quality, affordable APIs for historical and fundamental financial market data. Our customers range from individual investors and developers to fintech companies and startups looking to integrate clean and accurate financial data into their products. 7. What’s most exciting about your traction to date? What excites us most is how quickly we grew after pivoting from a B2C investment analysis platform to a B2B API-first company. We are bootstrapped, have reached a six-figure MRR and built a loyal client base by solving a real market pain point. 8. Where do you think your growth will be next year? We expect growth to come from expanding our data sources, improving API performance, and reaching new markets — especially in regions underserved by traditional financial data providers. In addition, we see strong growth potential through the integration of AI into our services, particularly in data checking and validation. This will not only improve data quality but also enable us to offer new, intelligent services that better meet our users' needs. 9. Tell us about your first paying customer and revenue expectations over the next year. Our first paying customer signed up during the launch weekend after we built a simple landing page. It was an early sign that we were solving a real problem. Based on our current trajectory, we anticipate sustained revenue growth throughout the next year. 10. What’s your biggest challenge? Our biggest challenge is managing the transition from a small, fast-moving team to a growing organization with evolving processes. As we scale, things change rapidly — which requires us to stay agile, maintain control, and constantly anticipate what’s coming next. We need to combine strong operational discipline with visionary thinking to stay ahead of market shifts and external events. For example, when the market is booming, subscriber numbers grow. But when the market slows down or crashes, interest in trading often declines due to psychological factors — people lose confidence or become discouraged. Navigating these fluctuations and keeping users engaged through both market highs and lows is a constant challenge we’re focused on. This startup founder interview template is based on HackerNoon Founder & CEO David Smooke’s ten questions for startup founders.
Would you like to take a stab at answering some of these questions? The link for the template is HERE. This startup founder interview template is based on HackerNoon Founder & CEO David Smooke’s ten questions for startup founders. This startup founder interview template is based on HackerNoon Founder & CEO David Smooke’s ten questions for startup founders. HackerNoon Founder & CEO David Smooke ten questions for startup founders Would you like to take a stab at answering some of these questions? The link for the template is HERE. The link for the template is HERE.

Stories, News & Mentions about FIG

Categories

Trending Topics

Stories, News & Mentions about FIG

Fig’s Stories on HackerNoon

Categories

Trending Topics