Full Stack Developer

Full-Stack Engineer, AI Products - Remote

About the role

Finagra builds software for agriculture — planning, field operations, and the data that connects them. We're early enough that the product surface is still expanding: what exists today is the first two systems, not the last two.

Right now that's Morpheus, which generates seasonal operations plans and SOPs from a corpus of agronomic source material using LLM agents and deterministic Python, and Trinity, the mobile-first execution system where agronomists approve those plans and field teams work them — daily checklists, evidence capture, deviations. More is coming: new products, new surfaces, and problems in this domain nobody has built software for yet.

You'll build across all of it — user-facing applications, the services behind them, and the AI features that make them worth using.

What you'll do

• Build product across the stack: TypeScript end to end (Next.js/React on the front, a hexagonal core and Postgres behind it), plus Python where the domain logic lives

• Build AI features that ship: agent pipelines, retrieval over domain corpora, structured generation validated against schema — not chat wrappers

• Work on output quality where it matters: prompts, evals, schema gates, and the judgement to know when a deterministic script beats a model call

• Take new products from nothing: greenfield services and apps as the roadmap opens up, alongside the systems already in production

• Own features end to end — scope, build, deploy, watch them in production, iterate

• Keep the domain honest: approval gates, audit trails and state machines here are product requirements, not plumbing

What we're looking for

• 3+ years shipping production web applications

• Strong TypeScript across frontend and backend; comfortable picking up Python

• Hands-on LLM integration in real products — you've dealt with context limits, latency, cost and non-deterministic output reaching real users

• Instinct for where a model belongs and where it doesn't. Half of good AI engineering is choosing not to use one

• Comfortable with typed contracts, schema validation and tests as the way work gets accepted

• Product sense: you ask who uses this and what breaks for them before writing code

• You want range. This is not one codebase for three years

Nice to have

• Structured generation, evals, or agent frameworks (we use the Claude Agent SDK)

• Postgres schema design, or hexagonal/ports-and-adapters architecture

• Vercel, Cloud Run, Docker, CI as a real gate

• Agriculture, commodity trading, or agri-finance domain exposure

• Work we can look at — repos, live products, demos

Why this is worth your time

• Genuinely remote. Not hybrid, not "remote-friendly, three days in." The company is built to work distributed — decisions happen in writing, so being in a room isn't how you get heard. Work from wherever you're actually productive, on a schedule that fits your life.

• Funded and growing, not scraping. We're backed by investor, with runway to build properly rather than ship whatever demos next quarter. You won't be rewriting your own work every six weeks because the company pivoted to survive.

• Ownership, not tickets. Small team, flat structure, no product-manager layer between you and the problem. You'll talk to agronomists and field teams directly, decide what to build, and ship it. The architecture decisions in this codebase were made by the engineers who write it — and yours will be too.

• An AI-native engineering environment. We don't just build AI features, we build with AI: our development process runs on machine-enforced gates from intent through spec, plan and PR, and agents are first-class contributors to the codebase. If you want to be genuinely good at engineering in 2027 rather than watching it happen, this is where that practice is being worked out.

• Real range, fast. Frontend, backend, domain modelling, LLM pipelines, infrastructure — inside one year, not one per role change. Very few companies this size will hand you that surface area.