Projects — Nisarg Bhatt

Multi-Modal AI

Mind Palace

Search your memories like you chat but with photos, screenshots, and moments. Mind Palace lets you describe what you're looking for in plain English and instantly finds the visual memory that matches. Built with privacy at its core, with active research into running everything on-device so your data never leaves your hands, and your memory stays with you.

CLIP LLaMA MongoDB FastAPI HNSW Nemotron

View project

Assistive Tech

Netrr

A lightweight wearable that gives visually impaired people a real-time understanding of the world around them — from recognizing familiar faces to describing entire scenes through voice. Works offline for everyday tasks and connects online for richer interactions. Co-designed with the Ahmedabad Blind People Association to solve real problems, not imagined ones.

TensorRT Edge AI Computer Vision Quantization

View project

Robotics

EcoBot

A self-driving robot that navigates public spaces, identifies waste, and picks it up — no human operator needed. It learns to coordinate its movement and arm control through trial and error, getting smarter with every run. Built in simulation as a proof of concept for scalable urban cleanup.

RL PyBullet SLAM A* Robotics

GitHub

Audio ML

Voice Biometrics Pipeline

Know who's speaking — even in noisy, real-world audio. This production system identifies speakers from their voice alone, and gets smarter over time by intelligently choosing which samples need human review. Less manual work, better accuracy, built for scale.

ECAPA-TDNN Active Learning Audio ML Python

Thout.ai

RAG Systems

Multi-Modal RAG Pipeline

Ask a question, get an answer grounded in your actual documents — not hallucinated. This system reads and understands large, messy document collections (PDFs, slides, reports) and pulls the right context to give accurate, source-backed answers every time.

RAG Embeddings LLMs Vector DB

Thout.ai

Agentic AI

6-Agent LLM Orchestrator

Tell it what you need in plain English — "create a sprint for the auth feature" or "summarize last week's standup notes" — and six AI agents coordinate behind the scenes to get it done across Jira, Notion, and 28 other tools. If something breaks, it fixes itself and tries again.

LLM Agents FastAPI Jira Notion

Thout.ai