I design and ship full-stack AI products end-to-end — frontend, backend, agentic systems — with the analytical rigour of a research scientist. MSc Chemistry, IIT Hyderabad. Self-taught engineer. Less than 6 months in. Already deployed.
Aswin · IIT Hyderabad · 2026
I'm an AI-native builder finishing an MSc in Chemistry at IIT Hyderabad (expected July 2026). I started teaching myself to code less than 6 months ago and have since shipped multiple production systems — deployed, live, usable.
My background in research gives me something most self-taught developers lack: the discipline to reason from first principles, design experiments, and question assumptions before writing a line of code.
I work at the intersection of product thinking and technical execution. I don't just build what's asked — I identify what's missing, design the right solution, and ship it fast using modern agentic workflows.
I'm looking for early-stage roles where I can build things that matter — AI products, developer tools, or anything that requires both analytical depth and rapid execution.
Architected a multi-stage agentic reasoning loop (Observe → Extract → Generate → Critique → Self-Correct) using Mistral-7B. The Critique step heuristically corrects LLM output before the user sees it — production-grade reliability thinking beyond basic API calls. Dual deployment: Gradio backend on HuggingFace Spaces, Next.js/TypeScript frontend on Vercel, plus a full terminal client with 256-colour ANSI mascot rendering and threaded animation.
Identified a product gap: existing Python resources used outdated syntax and passive formats that fail engagement-sensitive learners. Rebuilt the experience as a reward-driven, mission-based system with XP mechanics and a cyberpunk aesthetic. Shipped end-to-end: Next.js frontend on Vercel, FastAPI backend on HuggingFace Spaces, Supabase for persistent user state and progress tracking. 30 structured missions updated to modern Python 3 throughout.
Built a 5-stage agentic pipeline in 72 hours: JD parsing → candidate discovery → weighted scoring → conversation simulation → ranked shortlist. Designed scoring model from first principles: 60% match quality + 40% interest signal from a simulated 3-turn conversation. Implemented multi-model LLM routing with budget control and token tracking. Post-build, identified the CLI adoption barrier and proposed a v2 web SaaS architecture — product thinking beyond the engineering task.
I'm actively looking for entry-level roles in AI, full-stack development, or product. If you're building something interesting and need someone who ships fast and thinks clearly — reach out.