Reshinth Adithyan
Research Engineer at Stability AI · London
I'm a Research Engineer at Stability AI, based in London, on the Code.AI team — where I predominantly work on the post-training of LLMs; previously I worked on the large-scale pre-training of CodeLMs. I helped build the Stable Code suite of models and now do extensive research on LLM×RL post-training. Outside of work, I enjoy Impressionist painting and writing about music.
I was also a core member of Code.AI, an open-source community of researchers and engineers working on code intelligence and ML-aided code completion, where I was a major contributor to Code Clippy — one of the first open-source code language models.
Research
- LLM×RL post-training
- Code LMs and qualitative evaluation
- Interpreter-grounded code LMs
Selected Publications
Stable Code Technical Report
Stable LM 2 1.6B Technical Report
Patents
Method and system for translation of codes based on semantic similarity
Method and system for inferencing logic out of an application source
Method and system for extracting natural language elements embedded in application source code
Experience
2023 — Now
Research EngineerStability AI
Post-training of LLMs and LLM×RL research; previously large-scale pre-training of code models (Stable Code).
2022 — 2023
Research EngineerSaama Technologies
ML-aided program synthesis for clinical data transformations.
2019 — 2022
Research EngineerMastercraft Labs, TCS
ML-aided static analysis for COBOL — joined as a research intern in 2019.
Writing
Attempts to measure diversity of CodeLM generations
March 2023
Observing and measuring the diversity of code-LM generations — and the shortcomings of current methods.