Bengaluru
I’m Gautam — a Full Stack AI Engineer working on voice systems, multi-agent orchestration, and applied machine learning. I care about the part after the notebook: the latency budget, the failure path, the thing still standing when real users show up.
Bachelor of Technology, Information Technology
2023 – 2027
Aug 2026 – Present
Built a notification campaigns system with scheduled dispatch on GCP Cloud Run and batched email delivery, plus multi-document AI chat across up to 10 SEC filings, and cut API calls by 50–67% on high-traffic pages.
Nov 2025 – Jan 2026
Built a FastAPI outbound voice chatbot that holds real-time phone conversations, wiring Exotel telephony to Deepgram transcription, OpenAI reasoning, and Cartesia speech over persistent WebSockets.
2024 – 2025
Run the campus developer programme. Organised a national hackathon with 150+ participating teams, from sponsor coordination through to judging logistics.
Dev tooling
Git-like versioning for AI coding sessions.
Snapshots conversations into structured project memory, merges decisions across sessions into a persistent knowledge base, and generates briefings you can inject into Claude Code, Codex, Cursor, or Kiro. Architectural decisions, dead ends, and open questions survive the jump between tools — so long-running AI-assisted work stops rebuilding context from scratch.
Voice AI · Web3
Own your voice, earn on-chain.
A decentralised voice marketplace on Sui. Creators tokenise and sell custom voice models — stored on Walrus, registered on-chain, gated by blockchain access control — then deploy LiveKit voice agents that collaborate in real time. Purchased voices drive text-to-speech through Murf while royalties settle automatically.
ML · Forecasting
Hyperlocal air-quality forecasting for India.
An end-to-end platform, built as my major project for AI/ML roles. Live data from 250+ stations — AQI, weather, satellite fire detections — flows through a bronze/silver/gold TimescaleDB pipeline on Airflow; spatiotemporal models forecast AQI 72 hours out, and the Claude API explains why each spike happens with advice tuned to your health profile.
AI agents
Blueprints to auditable BOQs.
Turns tender PDFs and engineering drawings into accurate, traceable Bills of Quantities. It reconstructs the project as an interactive 3D model, then runs a five-agent pipeline to extract quantities, estimate cost, and flag bid risk — every number grounded in the source drawing. Weeks of manual estimation collapse to minutes, fully auditable.
AI infra
Any website into an MCP server.
Paste a URL and Lathe stands up a live MCP server in under a minute — provisioning from Anakin's Wire catalog, or crawling the site, inferring its schema, and hosting an endpoint. One config snippet later, assistants like Claude query the site as structured, real-time data. No scrapers, no APIs, no maintenance.
Agents · Memory
Tool-using chat with a memory.
An agentic conversation platform with persistent, thread-based memory. Built on LangGraph, it routes each query to the right tool — web search, weather, stock prices, a calculator — while keeping multi-turn history per thread in SQLite, so complex tool-augmented conversations stay context-aware across sessions.
Everything else lives on GitHub.
WinnerHyderabad, India
Codex Community Hackathon
WinnerBangalore, India
Google · Build with AI
2nd placeBangalore, India
Aptos Foundation
Runner-upBIT, Bangalore
National-level hackathon
If you’re building something in voice, agents, or applied ML — whether that’s a role, a contract, or just a good problem you want a second opinion on — email is the fastest path. I reply to everything that isn’t a template.
gautamsharma99067@gmail.com