I like working on AI because anything that takes us closer to a self-running society is good for us.
Lately that has meant spending time on building developer/agent-facing tooling, orchestration, review loops, and evals. Generally, I am comfortable across RAG, Agents, Developer Tools, and Multimodal Systems. And, I love Open Source.
Recent stuff:
- review-by-opp - keeps AI coding review honest by forcing findings to actually get closed. A Claude code plugin. (Made this about a month before codex came up with the same)
- modular-sdxl-upscale - tiled SDXL upscaling pipeline using Modular Diffusers, MultiDiffusion, and ControlNet Tile.
See pinned for more projects
open-source-contributions: selected fixes and reviews merged or adopted across external OSS projects including Huggingface (97th percentile of contributors to diffusers), MCP among many others
Codex for Open Source recipient for my OSS contributions to nanoclaw
Diffusers MVP for my contributions to HuggingFace Diffusers library
Applied AI, Software Engineering (Backend), AI agents, Language Modeling, Diffusion Models
| Deep Learning & Multimodal | PyTorch, MLX, Hugging Face, Diffusers, LLM/VLM fine-tuning (SFT, LoRA/QLoRA) |
| ML & Data | Python, scikit-learn, Pandas, NumPy, Polars, statsmodels, Gradient Boosting, MySQL, PostgreSQL, Cypher |
| Systems & Deployment | FastAPI, Node.js, AWS, MLflow, SQLAlchemy, GitHub Actions, Docker, Kubernetes, ONNX, Slurm, PyTorch FSDP, Tmux |
| Retreival, Agents & Serving | FAISS, Neo4j, LangGraph, OpenAI SDK, Claude Agent SDK, PydanticAI, vLLM, KubeRay |
| Languages | Python, Rust, Lua |
Open to interesting collaborations
Reach out to me to discuss Math, Neural Networks and Software Development akshankrithick305@gmail.com
If you're interested in open source AI or decentralized AI then we should definitely connect!
- Secured 172 in LSAT (best assessment for logical thinking skills and argumentative reasoning) {2024}
- 323 in GRE (assessment for academic readiness for graduate level studies) {2024}
- ~1300 elo on chess.com in rapid mode
- Dual-N-Back best level: 7
- Monkeytype: ~90 WPM with ~98% accuracy