Prarabdh Misra

AI / ML builder — retrieval-augmented systems, Kaggle competitions, and conversational robotics.

I build practical, end-to-end machine-learning systems: RAG assistants grounded in real research, competition-grade notebooks, and voice-driven robots. I care about shipping things that actually run, not just demos.

GitHub · Kaggle · LinkedIn · Hugging Face · Email

About

Hey, I’m Prarabdh — a high schooler (Class of 2029) in Johns Creek, GA, and I’m really into AI, machine learning, and robotics. Right now I like building things end-to-end: I made a RAG research assistant on Gradio, I work on machine-learning notebooks on Kaggle, and I built a little conversational robot that switches backends on its own if one goes down. What I really want next is a research internship where I can work on applied ML or RAG systems with an actual research group. Any mentorship, feedback on my projects, or a lead on a lab looking for a motivated student would mean a lot.

Areas of Focus

Retrieval-Augmented Generation — grounding LLMs in domain corpora with graph- and cost-aware retrieval strategies.
Applied ML & Kaggle — feature engineering, gradient boosting (LightGBM), and reproducible notebook workflows.
Conversational Robotics — hands-free, resilient voice agents with automatic multi-backend failover.
Edge ML / TinyML — quantizing and deploying models onto microcontrollers and single-board computers (ESP32 / Arduino, Raspberry Pi / Jetson).

Portfolio — Proof Points

Lab Assistant RAG — research-grounded, cited RAG assistant

A Gradio RAG chatbot that answers questions over a research lab’s publications with inline citations, expands results along a citation graph, and routes cheap vs. expensive models to save cost — with a grounding guard that declines unsupported questions. Provider-agnostic (Gemini / OpenAI / Anthropic / Hugging Face / Ollama), and open-sourced as a reusable template (MIT). Code on GitHub → · Live demo on Hugging Face →

Kaggle notebooks

Competition and EDA notebooks — exploratory analysis plus LightGBM baselines and iterations. Working toward Notebooks Grandmaster. View my Kaggle profile →

Reachy Mini “HARVI” — conversational robot

A hands-free voice robot built on the Reachy Mini platform. Talks in English with a dual-backend architecture that boots on Gemini (live voice + motion) and automatically fails over to a Hugging Face / Qwen backend if the primary is unhealthy — so it keeps working without intervention. Write-up: (link a blog post or repo)