Hi, I'm Tejas

I build AI systems that make it to production and to the YC stage. My work has ranged from a self-improving voice agent in a Y Combinator Hackathon, to real-time streaming infrastructure at Jio Platforms, to production ML pipelines that cut cloud costs. I care about the full stack fine-tuning vision and language models, building the distributed systems underneath them, and making sure the end product is clean and usable.

Let's connect:

FeaturedApril 2026

FreightVoice: AI Voice Agent for Freight Logistics

Carrier churn costs freight brokerages millions, but brokers can't monitor every call or catch dissatisfaction signals before carriers leave. FreightVoice fixes this with an AI voice agent that completes calls at sub-500ms latency, auto-scores agent performance after every conversation, and predicts churn risk earlier — built on Pipecat, NVIDIA Nemotron, Twilio, WebRTC, Cekura MCP, and XGBoost.

Mar 2026

instaMIND.tech

On-device video intelligence for security surveillance with <100ms P95 detection using C++/OpenCV and fine-tuned Gemma 3.

Apr 2026

Opusai.live

Local AI diagnostic assistant reducing Diabetic Retinopathy screening from weeks to <8s using fine-tuned PaliGemma 2 3B with QLoRA 4-bit.

Jan 2026

ClinSearch

Clinical NLP pipeline for natural-language search over 1,000+ NIH trials with LLM entity extraction, reducing missed matches by 30%.