About the role
Bellsoph is AI landlord software and UK property data. Our AI assistant reads tenancy agreements and certificates, answers questions about a landlord’s portfolio, files paperwork and talks to landlords in the app and on WhatsApp.
We’re looking for a full stack developer who is as comfortable in a Postgres schema as in a prompt. You’ll work across the whole product: the Next.js app, the Supabase database and the LLM pipeline that sits between them. We’re a small team, so what you build ships and gets used.
What you’ll do
- Build and ship features end to end in TypeScript, React and Next.js, with Postgres on Supabase.
- Improve the AI assistant: tool calling over a landlord’s records, streamed replies and approval steps before anything is saved or sent.
- Build retrieval that grounds every answer in the landlord’s own documents and data, so the assistant can show where an answer came from.
- Work on document extraction, reading tenancy agreements, certificates and receipts from PDFs and photos with OCR and vision models.
- Write evals and regression tests for prompts, tools and retrieval, and keep an eye on cost and speed for every reply.
- Keep landlords’ data safe with row level security, workspace access checks and guardrails so an agent never acts without permission.
- Connect the assistant to the channels landlords already use, including the WhatsApp Cloud API and email.
What we’re looking for
- Strong TypeScript, with production experience in React and Next.js or a similar framework.
- Solid SQL and Postgres, including schema design, migrations and access policies.
- Hands on experience building with LLM APIs such as OpenAI, Gemini or Claude: prompts, structured output, tool calling and streaming.
- A working understanding of RAG systems: chunking, embeddings, retrieval, reranking and how to check that an answer is grounded.
- Good testing habits. We use Playwright and run a full browser suite before every release.
- Clear written English and the confidence to take a short brief through to something shipped.
- Based in or near London.
Nice to have
- Experience with agents that take real actions, including approvals, undo and audit trails.
- WhatsApp Business or other messaging APIs.
- Python for data pipelines. Our property data pipeline is written in Python.
- Time spent in property, proptech or fintech.