Ayush Thakur
I’m an ML/AI Engineer with 6+ years of industry experience.
Currently I work as Manager, AI Engineer at Weights & Biases (now part of CoreWeave), where I’m building an agentic ecosystem for internal productivity. Earlier at W&B, I led open-source integrations that help ML practitioners build complete MLOps stacks, specialized in LLM applications and evaluation, created courses, and wrote extensively about what I’ve learned.
At the center of that is fixit, a background coding agent. The hard part isn’t creating a code diff—it’s getting an agent to single-shot bug fixes on our own private codebases, and figuring out what recursive self-improvement for agents actually looks like.
Alongside fixit, I’m building an internal benchmark to evaluate and optimize agent harnesses. SWE-bench and similar benchmarks aren’t indicative of real harness success, so I’m working toward an evolving benchmark that is.
Over the coming two quarters, I’m focusing on ML/AI research and actively thinking deeply about research ideas to work on.
News
- 2026 Won a solo Silver medal (rank 132/4729 teams) in the Kaggle Orbit Wars competition.
Experience
Manager, AI Engineering at Weights & Biases (CoreWeave)
February 2025 – Present
Leading a team building an internal coding agent and internal harness-evaluation benchmarks, and researching agent memory and self-improvement.
Machine Learning Engineer at Weights & Biases
August 2020 – January 2025
- Built MLOps integrations for Keras, OpenMMLab, Meta, XGBoost, and others
- Developed LLMOps integrations with OpenAI, Cohere, CrewAI
- Invested heavily in building evaluation for our RAG system, wandbot, and systematically improving it
- Built the Kaggle Ambassador program; became a Notebooks Master
- Authored multiple courses: LLM apps: Evaluation, RAG++: From POC to Production, and Training and Fine-tuning LLMs
Technical Author at Weights & Biases
January 2020 – August 2020
Wrote technical reports on ML concepts. Some were featured in Two Minute Papers.
Recognition
- Google Developer Expert in Machine Learning (2022–2025)
- Kaggle Notebooks Master
Education
B.Tech in Electronics and Communication Engineering
Netaji Subhash Engineering College, Kolkata (2016–2020)
Failed to crack JEE. Secured 91.56% in 12th boards and 10 CGPA in 10th board. Did my schooling from Notre Dame Academy.
Interests
- Robotics — My early work was on autonomous robots, which gave me a foundation in systems thinking that I still bring to AI. I like tinkering with hardware and 3D printing.
- Kaggle — I compete and share notebooks as a Kaggle Notebooks Master.
Why “Enginear”? When I was 10, my class teacher asked what I wanted to become when I grew up. I said “Engineer.” She asked me to spell it—I spelled it wrong, and she said, “You can’t even spell engineer, how will you become one?” Well, I am one now, and in a very high percentile category at that. Don’t let your teachers define what and who you can be.