|
I'm a Robotics undergrad who believes the best code is the kind you don't have to explain. I enjoy building things that sit at the intersection of software and the physical world from AI systems to robots that run across your desk. I care about clean systems, minimal interfaces, and not making future-me suffer. Currently deep in robotics, LLM tooling, and anything that blurs the line between software and the real world. Always learning in public. Always shipping. |
|
|
Connect with me:
|
Languages & Frameworks
AI / ML
Backend & Infrastructure
Tools
 Click Here
A Vision-Language-Action system for the SO-101 robotic arm. Combines MuJoCo physics, Groq (Llama 3) for planning, and Gemini Vision for real-time perception. Achieves ~80% plan success rate on pick-and-place tasks with sub-200ms LLM inference.
A fully local, privacy-first voice assistant built for Apple Silicon. Uses Whisper for speech recognition and runs Gemma 3 4B (4-bit quantized) via MLX — no cloud calls, no data leaving your device. Wake word: "Travis".
An open reimplementation of o1-style chain-of-thought reasoning using Gemini's API and a sequential multi-agent architecture. Proposes and iteratively refines solutions through reflection at each stage — built to explore how far smaller models can go.
An experimental clone of deep web research tools like Grok-3, built with DeepSeek R1 via Groq and SerpAPI. Implements multi-source validation, semantic scoring, and research-paper-style output. Proof that meaningful research automation is achievable with open models.
A Streamlit app that pits LLMs against each other in a real-time roast battle. Supports any OpenRouter model, configurable temperature and token limits, and live token-by-token streaming. Built for fun — but it's good multi-agent streaming architecture under the hood.
you scrolled all the way down here.
you're one of us. come build with me → shivamshsr@gmail.com