I'm a final-year B.Tech CSE (AI & ML) student at BVDU DET, Navi Mumbai —
building AI systems that work in production, not just on paper.
Previously at DRDO R&DE(E), Pune, where I built a reinforcement
learning quadruped locomotion system using MuJoCo and JAX — teaching a robot to walk
from scratch with PPO.
I'm interested in LLMs, agentic pipelines, robotics, and local inference.
I document what I build as a tech content creator on YouTube and Instagram —
because if you can't explain it, you don't understand it.
Perplexity-style AI search agent with multi-turn ReAct loops, SSE streaming, and Qwen3-8B as the tool-calling backbone. DuckDuckGo + Serper.dev search integration with a dark minimal UI.
GitHub ↗Local AI image generation pipeline — FastAPI backend, llama.cpp LLM inference, and ComfyUI headless with JuggernautXL. Runs entirely offline on Apple Silicon. Zero cloud dependency.
GitHub ↗RL-based quadruped locomotion built during my DRDO internship. PPO agent trained in MuJoCo physics simulation using JAX and Stable Baselines3 — from random flailing to stable gait.
GitHub ↗Automated road-littering detection via traffic CCTV. YOLOv8 detects violations in real-time, PaddleOCR reads license plates, FastAPI + SQLAlchemy handles persistence and alerting.
GitHub ↗Local meeting transcription with speaker diarization via pyannote, ChromaDB RAG for Q&A over your meetings, and Ollama LLM integration. FastAPI + WebSocket for real-time output.
GitHub ↗Final year project — AI exam proctoring system. MediaPipe pose estimation, YOLOv8 object detection, CNN audio analysis, all fused into a single Trust Score metric per candidate.
GitHub ↗Built a reinforcement learning-based quadruped locomotion system from scratch. Designed reward shaping functions, trained PPO agents in MuJoCo physics simulation, and iterated on gait stability across uneven terrain profiles using JAX and Stable Baselines3.
Led AI/ML workshops, organized hackathons, and mentored junior developers. Built a hands-on curriculum around practical ML projects to grow technical depth within the chapter.