Founding Engineer

Prysm is building the AI underwriting engine for private credit, one of the fastest-growing, highest-stakes corners of global finance. Our agents do the work that today swallows the bulk of an analyst's week: reading deal documents, researching markets, pulling comparables, building the financial model, and producing structured, fully cited underwriting outputs with every number traceable back to its source in one click.

This is agentic AI pointed at a problem where being right matters, and being able to prove you're right matters just as much. Credit teams screen around 200 deals a year and close around 10. The other 95% of expert time disappears into pulling numbers out of broker PDFs, rebuilding the same financial models, and reconstructing "where did this valuation come from?" three weeks after the fact. The audit trail is buried in spreadsheets, and when a senior person leaves, years of market intuition walk out with them.

Prysm replaces that with agents that extract, research, and synthesise structured, fully cited analysis.

Our wedge is commercial real estate credit, a $10 trillion global market, which is document and judgement heavy and tailor‑made for agentic AI, but the engine we're building generalises across credit.

Built by practitioners. Our CEO spent 20+ years in CRE credit, most recently as Partner at one of the largest global private credit funds, and our CTO was a Big 4 Partner. You'd be joining a small, senior, deeply technical team as a true founding engineer.

The role

Full time, based in London with hybrid working. You'll work directly with the CTO across the entire Prysm stack: frontend, backend, and the agent layer that is the heart of the product. This is a genuine generalist role with an AI centre of gravity. In a typical month you might:

• Improve the document-parsing pipeline so messy, inconsistent broker PDFs come through clean, structured, and queryable.
• Push the frontier of what agents can do in financial analysis: building the modelling agents that project cash flows, size debt, and run sensitivities as part of how a deal gets underwritten.
• Build the eval system that scores agent output against analyst-reviewed ground truth.
• Optimise an agent: tune prompts, restructure the lead/subagent orchestration, improve the extraction flows, and measure the quality difference.

You won't be siloed. The whole point of this hire is breadth: someone who can own a feature from database migration to frontend rendering.

The stack you'll work in

We build with a modern TypeScript-and-Python stack and lean on best-in-class tools for the AI layer. Specifics will shift as the product and the ecosystem evolve; what matters is being comfortable across the whole thing:

• Frontend: A modern Next.js/React/TypeScript web app
• AI / Agents: LLM agent orchestration with Pydantic AI and MCP; lead-agent + subagent patterns and tool design
• Data & retrieval: PostgreSQL with vector and full-text search, embeddings, and reranking powering RAG over documents
• Document AI: LLM pipelines that turn messy PDFs into structured, queryable content

What we're looking for

5-6+ years of professional engineering experience. We care far more about range, judgement, and trajectory than years on a CV.

Must-haves

• AI agent expertise. Hands‑on experience with agentic AI harnesses, e.g. LangChain or Pydantic AI, and building, optimising and orchestrating agents.
• Strong Python. You write clean, well‑structured, production Python and are comfortable in an async, API‑driven codebase.
• Full‑stack confidence. You can hold your own on the frontend in TypeScript/React, and you're happy owning a feature across the whole stack.
• A fast, fearless learner. New framework, new SDK, new corner of the system: you pick it up and ship. Our stack will keep evolving and we need someone who's energised by that, not unsettled by it.
• Genuinely excited by AI. This isn't a buzzword for you. You're inspired by what agentic AI can do and you want to spend your days building it.
• AI-native in how you build. Our development workflow is deeply agentic. You live in tools like Claude Code, Codex, or Cursor (whatever your weapon of choice) and you've built real fluency in driving them well.
• Experience with retrieval / RAG, embeddings, vector search, or document‑extraction pipelines.
• Exposure to finance, real estate, or other domains where accuracy and auditability are non‑negotiable.
• A track record of shipping in small teams or early‑stage startups, where you owned outcomes end-to-end.
• Real money, real stakes. You're building AI that does verifiable, high‑value work on institutional credit decisions, not a demo.
• The AI is the product. Agents aren't a feature bolted onto a CRUD app; they're the heart of it. You'll spend your days on real agent engineering: orchestration, tool design, RAG, evals, and traceability.
• Founding ownership. A product that’s already live. The architectural calls you make will stick.
• Ground‑floor equity. Founding‑engineer equity alongside a competitive salary.