Senior Founding AI Software Engineer

This is a rare and exciting opportunity for a hands-on builder who wants to take ownership and build 0 to 1 at a VC-backed, fast-growing startup. Build with former FDA decision-makers to define the future of biotech industry standards.

You don't have to know FDA or biotech, but you believe in vertical AI solutions and want to push the boundary of what is possible. Work across our entire stack from document ingestion to agents and evals, build 0 to 1 and grow with a young team. We're looking for a Senior Founding AI Software Engineer who is passionate about solving challenging problems in a highly regulated industry.

Headquartered in San Francisco, Deffai is an AI-native VC-backed seed-stage startup building AI platform for simulating FDA regulatory intelligence, built together with former FDA reviewers. We help therapeutics companies and their investors anticipate regulatory outcomes.

Backed by top-tier tech investors in the SF bay area, with advisors from Git Hub, OpenAI, and Mercor.

You get market standard comp, equity, viable path to Head of Engineering or beyond.

You will work closely with 2 engineers on the team, build, ship quickly and shape our product vision.

• Evals & Benchmarks: Build the eval pipeline for model benchmarking and regression suites. Use them to gate prompt, model and agent changes, measure precision and recall against expert judgment, and benchmark our performance with frontier models.
• Document Ingestion & Retrieval: Build ingestion and retrieval over diverse, long, and complex regulatory documents (scanned PDFs, Word files, tables, slide decks): parsing, structure-aware chunking, hybrid search, reranking, context assembly under token budgets, and knowledge graphs. Keep every answer traceable to a cited source, with provenance and versioning.
• Agents: Build and operate our multi-agent review workflows as they move to a shared Lang Graph engine: tool use, memory, tracing, and handling cost, latency, and failures in long-running runs.
• Expert
  
  Collaboration:
  
  Work with former FDA experts to accelerate their regulatory review workflows. Turn their feedback into product changes and new evaluation cases.
• Build & Ship: Own architectural decisions that balance innovation, scalability, cost, and long-term maintainability in a high-stakes domain. Review AI-generated code for what a diff does not show: migration locks, environment boundaries, and tenant scope.
• Data Engineering: Work with the team on the data pipelines behind our product: document ingestion, background jobs, and the PostgreSQL schemas. Make every pipeline idempotent and observable.
• Infrastructure: Support the team on our AWS infrastructure:
  Terraform-managed ECS services, CI/CD, and separate development and production environments.
• Security & Tenant Isolation: Support the team with tenant isolation, role-based access control, encryption, audit logging, and data retention controls in every service that handles customer documents.

• Cracked: You love to build.
• Strong opinions: You have strong opinions and strong product taste, and you're not afraid of pushing back.
• Relentless: You are tough, can handle the speed and execute quickly.
• Execution-Oriented: You move fast and focus on solving real problems.
• Clear Communicator: You collaborate well across functions and push decisions forward.

• 5+ years building and operating production software, including owning the architecture of a production system
• Experience building evaluations against expert judgment, such as golden datasets and precision and recall, and using them to decide what ships
• Experience shipping LLM systems to production, such as retrieval over complex documents or agent workflows
• Daily use of AI coding tools, with the judgment to catch what they get wrong in migrations, infrastructure changes, and data-access code
• Strong Python and data engineering experience: schema design, zero-downtime migrations, and reliable pipelines and background jobs on a relational database
• Experience running production systems on a major cloud with infrastructure as code, CI/CD, and separate development and production environments

• Worked directly with domain experts (clinicians, lawyers, scientists) to turn their judgment into requirements and test cases
• Retrieval techniques: hybrid search, reranking, parsing scanned PDFs and tables, or knowledge graphs (GraphRAG or similar)
• Agent frameworks and runtimes:
  Lang Graph, MCP, Claude Agent SDK, or sandboxed agent execution
• Post-training…