careers · full-time
Senior AI Engineer.
Own the agent layer that runs against live customer data every day — and help shape the headless kernel the company is betting on.
the seat
Take direct ownership of the AI agent layer that makes the AI Intelligence Workforce more than an ERP — the conversational agent, the background workers, and the knowledge base that grounds them — and harden it into a product that runs unattended on live customer data, every day, without a human catching its mistakes after the fact.
at a glance
- Employment
- Full-time · PotionLabs
- Location
- Park Street, Kolkata (on-site)
- Start
- Immediate
- Openings
- 1 opening
- Compensation
- Discussed in the first conversation
what you own
Four surfaces, and you are accountable for all of them.
AI agent systems
- Build, extend and own the live agent layer: the conversational ops assistant, the background workers that read and act on order and procurement state, and the knowledge base that grounds their answers.
- Move AI touchpoints from “drafts a document when the deterministic path fails” toward agents that reliably act on their own, inside clear human-approval gates.
- Design for idempotent, retry-safe processing — the agents run continuously against a live production database, not a demo.
Platform engineering
- Build and maintain backend services (Python/FastAPI, direct Postgres access, no ORM) and the React/TypeScript frontend they support.
- Own features end to end — design, ship, then monitor in production — across both the transactional platform and the AI layer.
- Keep the engineering discipline intact: append-only audit logs, atomic numbering, wire contracts honoured across repos, CI gates green.
Architecture and direction
- Contribute directly to the headless kernel — the destination architecture where a new customer is read, not custom-built for.
- Make and defend real technical trade-offs: what stays deterministic today versus what should become AI-driven next, what generalises across customers versus what is one-off.
Business logic and team
- Understand our customers' business logic well enough to encode it — and be proactive about finding where it is wrong.
- Review and mentor the work of the AI Engineer Intern cohort; set the technical bar for what “production-ready” means here.
day one, you inherit
- The agent layer running against JD Jones & Co's live production data today.
- A direct hand in shaping the headless kernel — the architecture the company is betting its future on.
- Technical mentorship of the AI Engineer Intern cohort.
what we need from you
Experience
- 5+ years building and operating production backend systems — not tutorials, not side projects that never shipped.
- Hands-on experience building AI agents that take real actions in production: tool-calling, workflow orchestration, retrieval and knowledge grounding — not a chat wrapper around a model.
- Comfortable owning a live system: on-call judgment, idempotent processing, queues and workers, debugging a production incident at 11pm if one happens.
Technical skills
- Strong Python, production-grade — not “can write a script”.
- Strong SQL: schema design, complex queries, query optimisation, comfortable working directly against Postgres without an ORM.
- API design and backend architecture: REST, webhooks, background workers, event-driven pipelines.
- Comfort designing config-driven or schema-driven systems that generalise across customers, rather than hardcoding one customer's logic.
How you work
- You have shipped and operated an AI agent in production, and can talk concretely about where it broke and how you fixed it.
- Ownership mentality — handed an ambiguous operational problem, you come back with a working system, not a plan.
- Entrepreneurial: you spot gaps nobody assigned you, and move without waiting for permission.
- You can mentor less experienced engineers without slowing down your own output.
not required
- React/TypeScript familiarity — a plus, not a requirement. This is a backend and AI-systems role first.
who you are joining
PotionLabs.
We are an AI solutions company. We build AI-native systems that run the operational core of a business — not another dashboard bolted on top of one. We sit on top of the system a customer already has; their books stay in Tally or SAP.
Our first product, the AI Intelligence Workforce, is live in production at JD Jones & Co., a century-old manufacturer, where order entry went from about 8 hours to about 2 minutes. The headless kernel — the architecture where a new customer is read rather than custom-built for — is in build.
Manufacturing is our proving ground, not our ceiling. The team works from the Park Street office in Kolkata.
how to apply
Apply with what you have built and owned in production — systems, incidents and outcomes, not certificates. Degree, college and CGPA do not matter here; neither does a job title at a past employer.