Founding Engineer, AI & ML Systems

Filmore•Austin, TX

2d•Remote

About The Position

FilmoreAI is building the intelligence layer for the construction equipment industry, a multi-hundred-billion-dollar economy. Dealers currently manage the machine lifecycle across numerous disconnected systems and rely on tribal knowledge. FilmoreAI is creating a data and AI system to address this by building equipment domain-specific reasoning using a proprietary ontology. This system connects various data sources like ERP work orders, CRM opportunities, OEM telematics, UCC filings, auction results, and DMS transactions into a unified model of machines, customers, and dealer interactions throughout the lifecycle. The initial focus is on the aftermarket, where data is often the messiest and the potential for leverage is highest. The product is the data system itself, which includes parsing public information across all 50 states (UCC liens, construction projects, contractors, land development), normalizing telematics across OEM standards, and resolving entities across disparate systems.

Requirements

3+ years of experience with LLM APIs in production, including designing extraction schemas, building agentic workflows, writing evaluations, and reasoning about cost/latency/accuracy at scale.
Experience with modern agent stack components such as LangGraph (or equivalent typed agent framework), Pydantic AI, MCP, pgvector, golden-set evaluations, and prompt caching.
5+ years of strong Python experience, focusing on production-grade service and pipeline code with emphasis on async, typing, and packaging.
5+ years of SQL and Postgres experience, including schema design, migrations, query optimization, materialized views, and index strategy.
5+ years of cloud deployment experience, preferably on Azure (Container Apps, Blob Storage, Azure Database for PostgreSQL, Synapse Analytics), with experience shipping to production.
Experience handling messy real-world data, including inconsistent schemas, pagination edge cases, auth flows, dynamic JS-rendered pages, and document parsing.
Familiarity with modern data and agent stack components like Temporal, LangChain, LangGraph, Pydantic AI, pgvector, and MCP.
Ability to operate without direct supervision, scope problems, ship solutions, and identify incorrect problem statements.
Ability to navigate ambiguity, adapt to changing specifications, and identify signals of broken processes versus healthy startup velocity.
A bias towards shipping the smallest thing that proves a bet, starting with manual versions and building APIs only when justified.
Calibration and bias toward action, including admitting when knowledge is lacking, flagging incorrect data, and implementing guardrails for suspect LLM outputs.
A strong sense of mission alignment with helping dealers modernize their businesses.
Experience driving agentic IDEs (e.g., Claude Code, Cursor) as a primary loop for development.
Experience running agents in parallel across multiple worktrees, sessions, or branches.
Experience designing context for agents (files, schema, examples, acceptance criteria) rather than relying solely on clever prompts.
Experience orchestrating agents like services, with typed I/O, structured outputs, retries, tool registries, golden-set evaluations, and end-to-end observability.
Experience reasoning about the model layer in production, including tradeoffs for Opus, Haiku, Gemini, and OSS models, as well as routing, failover, prompt caching, and provider concentration risk.

Nice To Haves

Bonus: dbt and a modern warehouse (Synapse, Snowflake, Databricks).
GCP or AWS translation experience.
Experience with public/government/third-party data sources.
Experience with enrichment pipelines with fallback logic.
Experience with document extraction at scale.

Responsibilities

Design and ship the agent runtime, wiring 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.
Implement the staged trust ladder for write-back, progressing from read-only to human-approved to scoped autonomous actions, with all actions logged in the Postgres action ledger alongside reasoning traces and rollback paths.
Own the in-house Model Router, managing calls across Claude, GPT, Gemini, and OSS models, optimizing for cost, context, and capability.
Implement production-level features for the model layer, including prompt caching, failover strategies, provider concentration risk management, and evaluation-gated rollouts.
Build extraction and reasoning workflows for semi-structured and unstructured documents (filings, work orders, spec sheets) using vision and long-context models.
Design typed schemas with Pydantic, iterate on prompts as document formats change, and maintain explicit tradeoffs for cost, latency, and accuracy.
Build golden sets and evaluations to ensure confidence in shipping changes.
Migrate and extend the canonical schema, focusing on entity resolution across diverse data sources.
Maintain the OLTP plane in Azure Database for PostgreSQL (with pgvector) and the OLAP plane in Synapse, ensuring proper workload distribution.
Build and maintain Python ingestion pipelines for public, third-party, and partner data sources, handling real-world failure modes like rate limits, schema drift, and authentication.
Operate the stack on Azure, utilizing services like Container Apps, Blob Storage, Container Apps Jobs, and Temporal Cloud for orchestration and human-in-the-loop processes.