Headline

I build AI systems that watch, think, and act — then I build the teams that ship them | EM | Teachmint | IIT Bombay

About

I build AI systems that watch, think, and act — and the teams that ship them. Over 14 years I've gone from architecting high-scale distributed services to leading engineering teams to, most recently, shipping production computer vision and agentic LLM systems end to end: a platform that audits thousands of cloud-kitchen cameras in real time, an AI agent that watches live crowds and adapts evacuation plans on the fly, an ops assistant people run entire workflows through over WhatsApp. I'm equally at home writing the model-inference pipeline and running the team building around it. I take problems that don't have a clear path yet — and hand back something deployed, monitored, and actually used. Bangalore · usually building something.

Location

Bengaluru

AI stack

claude code

codex

langgraph

livekit

CrewAI

Figma Make

Work experience

Engineering Consultant — Rebel Foods (Aug 2025 — Current) at Rebel Foods, Aug 2025 - Present · 1 yr 1 mo, Bengaluru, Hands-on consultant building two greenfield AI products end-to-end for Rebel Foods' platform team. Internal Ops Chatbot (Feb 2026 — Current) WhatsApp + dashboard AI assistant for ops staff to query inventory and Google Sheets. WhatsApp Web → Node.js bridge → FastAPI → CrewAI + Gemini. Pluggable ACTIVE_AGENT swaps agents without redeploy. 3 CrewAI agents on Gemini 2.5 Flash: Default (Manager + Sheets Specialist over Sheets/DuckDB SQL), Simple, Inventory (Spark API + persisted multi-turn state for SKU/store resolution). Two-identity auth: org-restricted Google JWT + per-user OAuth for Sheets; tokens server-side only. Voice mode via LiveKit + Google STT/TTS worker reusing the typed-chat pipeline. React + Vite operator dashboard, Alembic-managed Postgres, Chroma RAG, DuckDB mirror of Sheets. Stack: Python 3.12, FastAPI, CrewAI, Gemini, Postgres, SQLAlchemy, Chroma, DuckDB, Node.js, React, Vite, LiveKit, GCS. IntelliCam — CV Audit Platform (Aug 2025 — Jan 2026) Audits cloud-kitchen operations from existing in-store CCTV. 3-service architecture on MongoDB + GCS: React → FastAPI → Pub/Sub → Brain (Python subscriber) → Mongo. Decouples UI from GPU-heavy analysis. Brain pipeline (Prefect): YOLO + ByteTrack → ReID (ResNet50/OSNet) PersonGallery for cross-frame tracking → ray-cast zone mapping on operator polygons → Gemini 2.5 Flash on Vertex AI with equipment-labelled reference frame. FastAPI: JWT auth, internal router for Brain callbacks, APScheduler-driven audits, Shinobi-controlled RTSP recording. React frontend: react-konva ROI/zone canvas + multi-step equipment-labelling flow. RTSP recorder: containerised Python on GKE via Helm; GCS upload + Pub/Sub trigger. Stack: Python, FastAPI, MongoDB, GCS, Pub/Sub, YOLO, ByteTrack, ReID, OpenCV, Prefect, Vertex AI, Gemini, Shinobi, React, react-konva, Docker, K8s, Helm.

Stealth Startup at Clean Energy Solutions, Jul 2024 - Nov 2024 · 5 mos, Remote, Currently working on a stealth-mode startup focused on pioneering clean energy innovations. Conducting deep research into sustainable technologies, energy efficiency solutions, and market opportunities. Building the foundation for a breakthrough product aimed at reducing carbon footprints and driving energy sustainability.

Engineering Manager at Teachmint, Apr 2022 - Aug 2023 · 1 yr 5 mos, Bengaluru, Karnataka, India, -Managed a 10+ member cross-functional engineering teams comprising mobile, FE, BE developers in designing and developing features like HRMS, Student Report Card, Certificate builder -Implemented agile methodologies with rigorous estimation exercises to reduce delay of feature releases by 80% -Hired, coached and promoted software engineers and QAs -Was awarded best EM while team was awarded best pod for delivery execellance -Trained engineering teams on sprint process such as planning, refinement, stand-ups, reviews, and retrospectives. Improved efficiency, collaboration, and agile adoption for streamlined project execution and timely deliverables. -As part of EEG team designed Tech doc templates, setup PR review process, maintained tech backlog and reviewed code and architecture of on going projects -Tech stack- Python, Flask, Android, Nodejs, Reactjs, Kotlin, Kafka, Mongo, GCP, Graphana

Lead Software Engineer at MoEngage Inc., Nov 2019 - Apr 2022 · 2 yrs 6 mos, Bengaluru Area, India, Brief:- InApp team owns a service that delivers campaigns to Android/iOS devices or to Web sites through SDKs As a lead engineer I have designed and developed multiple features ensuring on-time delivery with help of 4 member team. Some of them are- InApp Presets, InApp client, InApp integration with Flows, AB Testing and Control Group Developed and maintained InApp service which handles a concurrency of 120k RPM at p99 Setup different tools like newrelic, sentry, Graphana, redash monitor performance and application failures Tech stack- Python, Falcon, Java, Mongo, Kafka, radis, Kubernetes, AWS, Celery

Education

btech, Aerspace Engg, Indian Institute of Technology, Bombay

C.B.S.E.Modi Public School

Public links

github: https://github.com/wittyurchin

medium: https://medium.com/@purnachandra

purna-patro-portfolio.web.app: https://purna-patro-portfolio.web.app/