Info
Aktivitäten
2264 Follower:innen
-
Sergii Khomenko hat dies geteiltJoin us to hear what is new in #TensorFlow and learn something for your daily ML workSergii Khomenko hat dies geteilt#TFEverywhere2021 Its my honor to serve #tensorflow user group community as Azerbaijan, as speaker on cutting edge research in #Tensorflow. This event is scheduled as follow: (6th March) 5.45PM - Keynote . 6.00PM - What’s New in TensorFlow 2.4? Sergii Khomenko , #Stylight 6.40PM - Introduction to On-device Machine Learning Karthik Muthuswamy , #sap 7.20PM - NLP using Tensorflow HUB Luiz GUStavo Martins, from #Google 8.00PM - Zero to Hero with TensorFlow Josh Gordon , from #Google 8.45PM - Cutting Edge Research in TensorFlow , me :) , #IBAR 9.30PM - Superpowers with TensorFlow.js , Jason Mayes , from #Google. (+4GMT, baku time.) Special thanks for supporting this event by The International Bank of Azerbaijan. Specially : Nihat Şenyuva Mustafa Basarir Zaur Alakbarov Leyla Hajieva Zara Ibrahimzade and many others. Last but not least Nijat Zeynalov and #Tensorflow community and #google developers group. Event Link on Google: https://lnkd.in/eU6MgE9 Registration Form: https://lnkd.in/eZKaXJp Event link on Linkedin: https://lnkd.in/dHkNKPQ
-
Sergii Khomenko hat dies geteiltThanks for having me!Sergii Khomenko hat dies geteilt"As a #startup you have much more flexibility on the market than a larger company. You can move faster, you can play different roles, you have less strict rules. So be as creative as possible and leverage to the best your small stature to address big challenges and capitalize on opportunities."- Sergii Khomenko #knowledgepill from the #BeAI #preaccelerator Mentor Sessions for the 10 selected startups. Watch them live on December 8, #pitchday.
-
Sergii Khomenko hat dies geteiltSergii Khomenko hat dies geteiltArtificial Intelligence meetup @ Stylight HQ in Munich with Ines Montani (Core spaCy developer) and Thomas Viehmann (Core Pytorch developer). High-level conversation on ML applied to NLP. To keep batteries fully charged: Pizza + Beer + Chocolates. In this fantastic environment we proceeded discussing ML till almost 23:00hs. On an amazing Friday night! Kudos to Sergii Khomenko for making it happen. #fridaynightintech #ai #onlyinmunich #pytorch
-
Sergii Khomenko hat dies geteiltSuch a great weekend with amazing people. Thanks everyone involved! #gdesummit2019Sergii Khomenko hat dies geteiltKicking off Day 1 at the #GDESummit2019 with #experts from around the world and Director of Operations at SpaceX and Astronaut Garrett Reisman!
-
Sergii Khomenko hat dies geteiltA great opportunity to join an awesome team working with millions products!
-
Sergii Khomenko hat dies geteiltIt was a great honor for me to give a talk about Machine Learning and Open Source for such remarkable leaders! The audience was really welcoming and authentic, so we ended up talking for a really long time about quite different topics. Such events give me a completely different view on diversity and inclusion. I've got quite a few great ideas on how to improve for my communities. #diversityandinclusion #machinelearningSergii Khomenko hat dies geteiltWhat a fun weekend in Poland at the #WTMSummit #WAW. With talks from such inspirational people, such as Emma Haruka Iwao telling us about how she smashed a world record. Laughing with Eliška Čejpová learning about design sprints while planning a UFO landing party 🛸 👽 To an intro to machine learning and tips for contributing to open source projects with @Sergii Khomenko . We met such inspirational people in Warsaw this weekend thanks so much for inviting me to be part of such a fun event. 🦄 Looking forward to meeting you all again.
-
Sergii Khomenko hat dies geteiltIt was amazing to be able to ask questions directly and listen to the interesting stories by Jeff Dean, Turing award recipient David Patterson and Fernando PereiraSergii Khomenko hat dies geteiltA handful of students and ML engineers were lucky to meet Jeff Dean at the #io19 ML/AI meetup. His advice to us was what he personally values: 1. working on interesting problems 2. with great people that have complementary skills and strengths 3. on projects with impact. #machinelearning #deeplearning #bayarea #google #brain
-
Sergii Khomenko hat dies geteiltSergii Khomenko hat dies geteiltIncredible time with an amazing community at the TensorFlow Dev Summit #TFDevSummit Go #ML #GDE team
-
Sergii Khomenko hat dies geteiltInvest in Bavaria - The Business Promotion Agency of the State of Bavaria
Invest in Bavaria - The Business Promotion Agency of the State of Bavaria
Sergii Khomenko hat dies geteiltDon’t believe everything about Bavaria. Just believe the facts.Don’t believe everything about Bavaria. Just believe the facts.
-
Sergii Khomenko gefällt dasSergii Khomenko gefällt dasA couple days ago I gave Claude Code the ability to debug GPUs with RenderDoc. Now I taught it Apple's Metal ecosystem. I built a second skill — metal-ai-skill — that connects Claude Code to Apple's GPU toolchain: xctrace, Metal validation layers, shader compilation, .gputrace frame capture, and screenshot-based visual inspection. When I describe a rendering bug on macOS, Claude can: - Screenshot the app and actually see what's wrong (multimodal vision) - Capture a .gputrace frame and inventory every buffer, texture, and shader - Compile shaders with -Weverything to catch static issues - Read source code to trace data flow from CPU to GPU - Fix bugs and re-screenshot to verify — in a loop until it's correct The interesting challenge: Metal's CLI tooling doesn't give you buffer data from programmatic captures (unlike RenderDoc's rdc-cli which gives you everything). So I built an autonomous debugging workflow — Claude gathers signal from multiple sources in parallel, checks what data is actually available, and falls back to source code analysis when GPU data isn't there. Every path leads to a diagnosis. In the demo, I give Claude a Metal cube renderer with 4 intentional bugs: collapsed geometry, flipped Y axis, swapped color channels, and zeroed alpha. Claude screenshots the broken output, reads the shaders and Swift code, diagnoses all 4 bugs, fixes them, rebuilds, and verifies the output — all autonomously. The RenderDoc skill taught me the pattern. The Metal skill taught me the fallbacks. The real insight: when you can't inspect GPU state directly, combine enough indirect sources and you still get there. Open source (MIT). GitHub: https://lnkd.in/g8n8qb6J Video: https://lnkd.in/gryAY4YR #ClaudeCode #GPU #Metal #Apple #GraphicsProgramming #AI #GameDev #OpenSource
-
Sergii Khomenko gefällt dasSergii Khomenko gefällt dasOpus 4.6 is incredible at visual thinking! Just wrapped Excalidraw into an MCP App - there's something oddly satisfying about watching Claude draw diagrams live, stroke by stroke. In the demo, it sketches a Raspberry Pi 5 board based on web research alone. GitHub link in comments
-
Sergii Khomenko gefällt dasSergii Khomenko gefällt dasThe Coding Machine is dead. My entire professional identity that served me well since I was 12 years old has been replaced by AI in the span of a year. Instead of being defensive or delusional, I had to face reality and move on. I wrote up three archetypes that have replaced the Coding Machine. These are the new role models engineers aspire to and work towards. 12 year old Michael of today will aspire to be an Editor, Orchestrator, or Multiplier. I discuss my thoughts on the popular Path to Staff substack. https://lnkd.in/dv4RwaYj
-
Sergii Khomenko gefällt dasSergii Khomenko gefällt dasHuge. Someone just connected Claude to Blender. You can now do 3D modeling with text prompts instead of clicking through menus. 𝗛𝗼𝘄 𝗶𝘁 𝘄𝗼𝗿𝗸𝘀 BlenderMCP creates a socket server inside Blender. Claude sends commands through the Model Context Protocol. The integration handles object creation, materials, scene manipulation, and Python code execution. 𝗪𝗵𝗮𝘁 𝘆𝗼𝘂 𝗰𝗮𝗻 𝗯𝘂𝗶𝗹𝗱 Complete 3D scenes from text: "Create a low-poly dungeon with a dragon guarding gold" "Make this car red and metallic" "Add studio lighting, point camera isometrically" ↓ 250k+ AI developers read this newsletter to stay up to date with AI: AlphaSignal.ai
-
Sergii Khomenko gefällt dasSergii Khomenko gefällt das📈 This OpenAI Chart on Reddit, Inc. It looks interesting. But as the top comments immediately called it, it's a bit let's say misleading (no y-axis scales) The more likely real story, job openings started dropping in 2022, the Fed hiked interest rates (anti infaltion), big tech over-hired during COVID and course corrected etc. So it looks dramatic, but its likely 95% fundamentals and 5% AI impact? So AI is hyping the market but not the economy. Correlation doesn't equal causation (yet). Still makes for a spicy chart! via Neuron
-
Sergii Khomenko hat darauf reagiertSergii Khomenko hat darauf reagiertHappy to share that today I’m joining Anthropic 🙌 Feels like everything I’ve done has been leading to this moment. Looking forward to what’s ahead!
-
Sergii Khomenko hat darauf reagiertSergii Khomenko hat darauf reagiertI never imagined I’ll get to study at Stanford, let alone at the Graduate School of Business with the smartest leaders from around the world. Today I’m receiving my Masters degree from the Stanford GSB with distinction (Robert Joss scholar — top 10% of the class). Feels unreal. I’m proud to display my national Ukrainian flag as I’m receiving this advanced degree. Like every Ukrainian-born proudly representing my home country at the international stage. The last time I felt this proud was in 2008 when our Lviv National University team for the first time in history won gold medals for Ukraine at the ACM international programming competition. Thank you, Stanford! Glory to Ukraine!
-
Sergii Khomenko hat darauf reagiertSergii Khomenko hat darauf reagiertGood summarization prompt --- 1.) Analyze the input text and generate 5 essential questions that, when answered, capture the main points and core meaning of the text. 2.) When formulating your questions: a. Address the central theme or argument b. Identify key supporting ideas c. Highlight important facts or evidence d. Reveal the author's purpose or perspective e. Explore any significant implications or conclusions. 3.) Answer all of your generated questions one-by-one in detail. https://lnkd.in/gMK7ZTPc ↓ Are you a developer? Check out https://AlphaSignal.ai to get a daily summary of breakthrough models, repos and papers in AI. Read by 200,000+ devs.
-
Sergii Khomenko gefällt dasNo matter how smart an LLM or RAG pipeline is, if the source documents are incorrect, the system will be just as good as the basic pipeline shown in slide 1. I can't help but roll my eyes whenever someone says, "Let's sync our SharePoint to a vector index and let Copilot run." If I had a penny for every time I hear this.... I'd be a person with a lot of coins. We at Shelf take a pragmatic, realistic approach to this problem. Focus on the quality of the source knowledge fed into AI. LLMs are rapidly evolving and will soon catch up. Invest in your data and knowledge.Sergii Khomenko gefällt dasMost people do not look beyond the basic RAG pipeline, and it rarely works out as expected! RAG is known to lack robustness due to the LLM weaknesses, but it doesn't mean we cannot build robust pipelines! Here is how we can improve them. The RAG pipeline, in its simplest form, is composed of a retriever and a generator. The user question is used to retrieve the database data that could be used as context to answer the question better. The retrieved data is used as context in a prompt for an LLM to answer the question. Instead of using the original user question as a query to the database, it is typical to rewrite the question for optimized retrieval. Instead of blindly returning the answer to the user, we better assess the generated answer. That is the idea behind Self-RAG. We can check for hallucinations and relevance to the question. If the model hallucinates, we are going to try again the generation, and if the answer doesn't address the question, we are going to restart the retrieval by rewriting the query. If the answer passes the validation, we can return it to the user. It might be better to provide feedback for the new retrieval and the new generation to be performed in a more educated manner. In the case we have too many iterations, we are going to assume that we just reach a state where the model will apologize for not being able to provide an answer to the question. When we are retrieving the documents, we are likely retrieving irrelevant documents, so it could be a good idea to filter only the relevant ones before providing them to the generator. Once the documents are filtered, it is likely that a lot of the information contained in the documents is irrelevant, so it is also good to extract only what could be useful to answer the question from the documents. This way, the generator will only see relevant information to answer the question. The assumption in typical RAG is that the question will be about the data stored in the database, but this is a very rigid assumption. We can use the idea behind Adaptive-RAG, where we are going to assess the question first and route to a datastore RAG, a websearch or a simple LLM. It is possible that we realize that none of the documents are actually relevant to the question, and we better reroute the question back to the web search. That is part of the idea behind Corrective RAG. If we reach the maximum of web search retries, we can give up and apologize to the user. Here is how I implemented this pipeline with LangGraph: https://lnkd.in/g8AAF7Fw
Berufserfahrung und Ausbildung
-
Meta
Gesamte Berufserfahrung von Sergii Khomenko anzeigen
Jobbezeichnung, Beschäftigungsdauer und mehr ansehen.
oder
Bescheinigungen und Zertifikate
Ehrenamt
-
Google Developer Expert in Machine Learning
Google
–Heute 8 Jahre 6 Monate
Ausbildung
Conducting lectures and workshops, speaking at conferences on topics of Data Science, Machine Learning, Deep Learning and related
ML Kickstarter Warsaw - Mentored a Machine Learning kickstarter workshop for selected startups
Talks:
27-09-2018, Zagreb, Croatia - Back To The Future: Building amazing products easily using TensorFlow
16-11-2018, Jurmala, Latvia - Mobile first to AI first: Making mobile apps smart with Machine Learning -
Founder, Organiser
Hacking Machine Learning
–Heute 9 Jahre 3 Monate
Wissenschaft und Technologie
Munich Applied Machine Learning community.
Focusing on making people familiar with machine learning and hands-on knowledge.
Currently, meetup has 2200+ members and hosted 35+ events with speakers from a variety of local, as well as top-tier companies like Google, Twitter, Facebook, etc. -
Munich Gophers - Golang GDG
Google
–Heute 12 Jahre 1 Monat
Wissenschaft und Technologie
Munich community of Golang developers. Regular meetings, community management.
http://www.meetup.com/Munich-Gophers-Go-User-Group/ -
Stage manager
Berlin Buzzwords
–Heute 10 Jahre 11 Monate
Wissenschaft und Technologie
Stage managing: introducing speakers on the main stage, Q&A sessions
Helped the conferences of 2015 and 2019 to make happen.
Veröffentlichungen
-
The optimization engineering computations in Microsoft Office: regression analysis with universal selection method
Intellectual systems of decision-making and problems of computational intelligence
The subject of the article is methods of optimization engineering computations in Microsoft Office. An algorithm for optimization, which is based on moving main calculations into compiled dynamic library, is displayed here. Also math methods for optimizations calculation regression analysis in way of using universal calculation method are considered here. In this way we can significantly reduce the number of calculations and increase speed of algorithm. Quantitative results of the coefficient…
The subject of the article is methods of optimization engineering computations in Microsoft Office. An algorithm for optimization, which is based on moving main calculations into compiled dynamic library, is displayed here. Also math methods for optimizations calculation regression analysis in way of using universal calculation method are considered here. In this way we can significantly reduce the number of calculations and increase speed of algorithm. Quantitative results of the coefficient usage effectiveness are given for the described method.
Andere Autor:innen -
-
Методи розпізнавання облич в растрових зображеннях
Праці конференції “ Технікуми і коледжі на шляху розвитку України у ХХІ столітті ”
Projekte
-
Personal Data Visualisation
-
Projekt anzeigenA weekend project in order to visualise check-ins from Foursquare with R, Shiny and Leafletjs. Added docker image for ease of test. As PoC added Tableau visualisation.
Results of the project were referenced in next talks:
Munich Quantified Self - Show and tell, Jan 13, 2015
Munich Tableau User Group - Tableau & R – personal data visualisation for fun, Jan 22, 2015
Munich Developer Camp - Personal Dataviz / Quantified Self made easy, Apr 19m 2015
Sprachen
-
Ukrainian
Muttersprache oder zweisprachig
-
English
Verhandlungssicher
-
German
Gute Kenntnisse
-
Spanish
Grundkenntnisse
Organisationen
-
Wearable Data Hackathon
Organiser
–HeuteThe Wearable Data Hack Munich 2015 is the first hack day on wearable tech applications, data and design. http://wearabledatahack.com/ https://www.heise.de/make/meldung/Wearable-Hackathon-Sachen-machen-mit-Sensoren-2720486.html https://www.youtube.com/watch?v=X-i_-hzMjuc
-
Tableau User Group München
Co-Organizer, Speaker
–Heutehttp://www.meetup.com/Tableau-User-Group-Munchen-TUGM/
-
Munich Gophers - Golang GDG
Co-Organizer, Speaker
–Heutehttp://www.meetup.com/Munich-Gophers-Go-User-Group/
