What We’re Building

FilmoreAI is the intelligence layer for the construction equipment industry — a multi-hundred-billion-dollar economy where dealers manage every stage of the machine lifecycle (acquisition, financing, utilization, service, trade-in, disposition) on data that lives in a dozen disconnected systems and a thousand reps' heads.

We're building the data and AI system that fixes that. We’re building equipment domain specific reasoning using a propietary ontology that connects ERP work orders, CRM opportunities, OEM telematics, UCC filings, auction results, and DMS transactions into a single canonical model of every machine, every customer, every dealer interaction across the lifecycle. Aftermarket, where dealers earn the majority of their profit on tribal knowledge is where the data is messiest and the leverage is highest, so it's where we lead.

The data system is the product. We parse public information across all 50 states including UCC liens, construction projects, contractors, and early land development. We normalize telematics across OEM standards. We resolve entities across systems that have never spoken to each other.

What You’ll Work On

• Agent infrastructure. Design and ship the agent runtime. Wire agent templates to live data through the canonical schema and the internal MCP tool registry. Build workflows in LangGraph with typed I/O, structured outputs, retries, and observability versioned and tested like backend services, not prompt notebooks. Implement the staged trust ladder for write-back — read-only → human-approved → scoped autonomous with every action logged in the Postgres action ledger alongside the reasoning trace and a rollback path.

• Model layer & routing. Own the in-house Model Router across Claude, GPT, Gemini, and OSS. Make the live calls on when Opus is worth the cost, when Haiku is enough, when Gemini’s long context is the unlock, and when an OSS model is the right call. Prompt caching, failover, provider-concentration risk, eval-gated rollouts the model layer is a production system and you operate it like one.

• LLM-powered extraction & reasoning. Build extraction and reasoning workflows over filings, work orders, spec sheets, and other semi-structured or unstructured documents using vision and long-context models. Design typed schemas with Pydantic, iterate on prompts as document formats change, and keep the cost / latency / accuracy tradeoff explicit including when an LLM is the wrong tool and a parser or regex is. Build the golden sets and evals that let you ship changes with confidence.

• Canonical data model. Migrate and extend the canonical schema. Entity resolution across dealer systems, telematics, public records, transaction histories, and third-party data is a primary, ongoing problem here and the schema is the compounding moat for every agent we build on top. Maintain the OLTP plane in Azure Database for PostgreSQL (with pgvector) cleanly separated from the OLAP plane in Synapse; know which workloads belong where and why.

• Data pipelines & cloud orchestration. Build and maintain Python ingestion across public, third-party, and partner sources — government registries, filing systems, geospatial APIs, permit and contract systems. Handle the real-world failure modes: rate limits, schema drift, auth flows, JS-rendered sites with Playwright. Operate the stack on Azure — Container Apps, Blob Storage, Container Apps Jobs, Temporal Cloud for long-running jobs, approval queues, and human-in-the-loop write-back paths.

The Stack

Data & Pipelines - Python, Playwright, public APIs; Airbyte + custom Python connectors; Databricks, MSFT Fabric, Cloud SQL Postgres with pgvector

Reasoning & Agents - Model Router across Claude, GPT, Gemini; LangGraph/LangChain for agent workflows; internal MCP tool registry; Postgres action ledger

Cloud & Backend - MSFT Azure, Temporal Cloud (orchestration + human-in-the-loop), Python + FastAPI, Terraform, GitHub Actions, Secret Manager

Delivery & Observability - Twilio (SMS), native CRM APIs, Datadog, LangSmith

What We’re Looking For

• LLM APIs in production (3+ yrs). You’ve shipped real systems with Anthropic, OpenAI, or Gemini designed extraction schemas, built agentic workflows, written evals, reasoned about cost/latency/accuracy at scale. You can tell a war story about a model regression you caught before the customer did.

• Modern agent stack. LangGraph (or equivalent typed agent framework), Pydantic AI, MCP, pgvector, golden-set evals, prompt caching. We don’t expect all of these we expect you to learn the ones you don’t and have opinions about the ones you have.

• Strong Python (5+ yrs). Production-grade service and pipeline code other engineers can read, extend, and trust six months later. Async, typing, packaging the boring parts done right.

• SQL & Postgres (5+ yrs). Schema design, migrations, query optimization, materialized views, index strategy. You read EXPLAIN plans without flinching. Bonus: dbt and a modern warehouse (Synapse, Snowflake, Databricks).

• Cloud deployment (5+ yrs). Azure preferred (Container Apps, Blob Storage, Azure Database for PostgreSQL, Synapse Analytics); GCP or AWS translate. You’ve shipped to production, not just dev.

• Messy real-world data. Inconsistent schemas, pagination edge cases, auth flows, dynamic JS-rendered pages, document parsing. You’ve debugged a scraper at 2am because a vendor changed their HTML.

• Modern data + agent stack familiarity. Temporal, LangChain, LangGraph, Pydantic AI, pgvector, MCP. We don’t expect all of these — we expect you to learn the ones you don’t.

• Strong plus: public/government/third-party data sources, enrichment pipelines with fallback logic, document extraction at scale.

• You operate without supervision. We hand you a problem, not a ticket. You scope it, ship it, and tell us when we got the problem statement wrong.

• You navigate ambiguity. The spec changes mid-week, the data is weird, and the customer feedback contradicts the design doc. You know when that's healthy startup velocity and when it's a signal something's broken.

• You ship the smallest thing that proves the bet. Manual version first. Build the API only when it earns its place. Walk away from problems that don't move the dealer's P&L.

• You're calibrated and bias toward action. When you don't know, you say so. When the data is wrong, you flag it. When an LLM output is suspect, you don't ship without guardrails. Then you keep moving.

• You care about why this exists. Dealers run their businesses on tribal knowledge and relationships. We're building the platform layer to help them modernize without implementing. If that mission doesn't pull you forward, the rest of this won't.

AI-Native

• You drive agentic IDEs as your primary loop. Claude Code, Cursor, or equivalent — not autocomplete, full agent sessions. You give the agent a problem, the right context, and the constraints, then review its work like a tech lead reviewing a strong junior. You know when to let it run and when to take the keyboard back.

• You run agents in parallel. Multiple worktrees, multiple sessions, multiple branches in flight — one agent migrating a schema, another writing tests, another drafting docs. You've adapted your planning, review, and merge discipline to a world where throughput isn't bounded by what one human can type.

• You design context, not prompts. You know an agent with the right files, schema, examples, and acceptance criteria does excellent work, and one with a clever prompt and no context does not. You write CLAUDE.md / agent specs / project rules the way you'd write a runbook — because you'll run them a hundred times.

• You orchestrate agents like services. Typed I/O, structured outputs, retries, tool registries (MCP), golden-set evals, end-to-end observability. LangGraph workflows are version-controlled, tested, and instrumented like backend services. Not prompt engineering. Software.

• You reason about the model layer in production. When Opus is worth the cost, when Haiku is enough, when Gemini's long context is the unlock, when an OSS model is the right call. Routing, failover, prompt caching, provider concentration risk — tradeoffs you've made for real, not in theory.

Compensation | To Apply

• Small team, direct founder access, decisions get made fast. Async-first — clear PR descriptions and run summaries, no standup theater. Every line of code accumulates proprietary data or makes the reasoning layer smarter; you’re expected to apply that filter too.

• Competitive pay + equity as early employee. Open to contracting and scope / hours if preferred.

To apply: email [email protected] with
(1) one or two data systems you've built — what you were solving, what made it hard, how you handled it
(2) how you've used LLMs in production with real data or real users — not demos
(3) the messiest data source you've worked with and what it taught you