Senior Machine Learning Engineer building agentic AI systems : RAG pipelines, emotionally aware multi modal systems, multi-agent frameworks, memory-aware assistants, and the infrastructure to evaluate whether agents actually behave correctly. I open-sourced AgentRubric, a trajectory-scoring framework for LLM agents, and Forge, a minimal GRPO fine-tuning engine for LLMs.

I'm drawn to Computational Neuroscience + Reinforcement Learning + JAX as a lens on intelligence itself, studying how the brain learns and generalizes with so little data, and bringing those principles into ML. I document that journey daily in JAX-RL-Lab, and turn the same curiosity into hardware and robotics: a game controlled by my own brain signals, SilentVoice, a sign-language-translating glove (open-source Instructable), and IRIS, a 3D-printed animatronic eye system I built and control with inverse kinematics and real-time hand tracking.

Outside of ML, I blend cinematography and tech. I directed the world's only short film on Data Structures and Algorithms, I build electronics + PCB + Neuroscience projects, 3D print and teach DSA on YouTube.

Currently: leading a Langfuse-native trajectory level LLM agent evaluation framework at SolarWinds (370+ eval scenarios), building INDEX 0, studying computational neuroscience + EEG applications, spoke on AI Safety at Google Development Fest to 700+ attendees.

Reinforcement Learning · Computational Neuroscience · LLM Agents & Evaluation · RAG · Transformers · JAX · Computer Vision · Speech · Applied Mathematics

A desktop application for DSA, ML/AI, electronics, and system design, built and shipped solo. Scaled to users across countries, and free for everyone, always.

Runs entirely on your own machine, no cloud dependency, account, or subscription. Bring your own model, Anthropic, OpenAI, LM Studio, Llama.cpp, or a fully local and free Ollama + Gemma setup. Includes local DSA coaching, guided mock interviews with Zero, your friendly fox coach, brain-capacity games, and a lecture gallery spanning AI/ML, electronics, and system design, all in one place. Because it needs no internet after setup, it's also reaching learners in parts of the world with little or no reliable connectivity.

Metric	Value
DSA problems	475
ML/AI lectures	114
Electronics lessons	74
System design lessons	102
Model providers	Anthropic · OpenAI · LM Studio · Llama.cpp · Ollama (Gemma)
Platforms	macOS · Windows · Linux
Downloads
Countries

www.index-0.in

• Forge: a minimal GRPO (Group Relative Policy Optimization) fine-tuning engine for LLMs, the algorithm behind DeepSeek-R1. Published on PyPI.
• AgentRubric: LLM agent trajectory evaluation across 5 metrics (tool accuracy, argument fidelity, efficiency, completion, failure taxonomy). Published on PyPI.
• IRIS: gesture-controlled animatronic eyes - a 3D-printed dual-eye robot on an Arduino Uno with two modes: autonomous lifelike idle motion via inverse kinematics, and real-time hand tracking (OpenCV + MediaPipe) where the eyes follow your hand and blink when you clench a fist.
• tinytutor: an accountability layer for AI-assisted coding. Plugs into Claude Code and quizzes you on the code it just wrote before letting you keep building, so you can actually explain your own project. Open source, published on npm.
• JAX-RL-Lab: daily JAX + RL implementations, from grad/jit/vmap to Q-learning, DQN, REINFORCE, and Actor-Critic.
• SilentVoice: sign-language-translating glove combining embedded sensors with ML gesture classification, published as an open-source Instructable.
• MoodMap: a multi-modal emotion-aware AI system, built around a self-trained audio emotion classifier, with video-based emotion recognition and speech-to-text support, aimed at giving AI a genuine sense of emotional context.