Role: Jr. Product Engineer (AI Systems and Conversation Architecture)

Location: Worli,Mumbai

Type: Full time

About the company

Owners delegate tasks through a web app or through Maya, the household concierge. Staff get structured WhatsApp instructions in Hindi, Marathi, English or Hinglish, complete tasks with photo confirmation, and every verified completion goes into an audit trail. The product is about 80% feature-complete, and the remaining 20% is the hard part.

What you will own

- Conversation architecture: routing, slot management, retries and interruption handling. A staff members photo confirmation always takes precedence over any open conversation state.
- Household memory layer: data model, retrieval and context assembly, so Maya remembers each households preferences, people, routines and history.
- Insight engine: classifying delight, friction and quit moments from analytics and audit logs, plus a weekly founder digest.
- Verification-loop reliability: queues, idempotency, webhooks and observability across the WhatsApp gateway, scaling from 27 to 50+ households.
- LLM layer: prompt architecture, cost-tiered model routing, eval sets and regression discipline.
- Technical leverage: code review and mentoring for a junior developer, and absorbing work from the external build partner.

First 90 days
- Support the existing dev cycle while stabilising each module.




- Ship the classifier-first routing refactor from an existing handover brief. This restores photo-completion precedence and removes the conversation-trap bug class.
- Stand up an eval harness for Mayas core intents.
- Take ownership of the WhatsApp gateways reliability: retries, dedupe and alerting.

Tech stack FastAPI and MongoDB (backend), React 19 PWA, NestJS WhatsApp gateway on a business service provider, Claude via LiteLLM with cost-tiered routing, and a VPS deployment.

Who you are

- 2 to 3 years as a product engineer or full-stack developer, with at least one LLM feature shipped to real production users.
- Able to describe your eval and regression setup for LLM features unprompted.
- Experience with messaging infrastructure such as WhatsApp Business, Twilio or chat systems, including webhooks, queues and idempotency.
- Experience as an early engineer (one to five) at a small company, comfortable working from written specs and phase gates.
- Restraint: keeping the concierge strictly scoped to household management is a product decision, not a limitation to engineer around.
- Hindi fluency is a meaningful plus.

Not for you if
- You want to train or fine-tune models. This is LLM application engineering only.
- You need a manager checking in daily. The CTO is fractional and the founder is non-technical day to day.

📌 Jr. Product Engineer (Mumbai)
🏢 Autowhat Ai And Automations
📍 Mumbai