Agentic AI Engineer

Quant Pilot is 3

Commas' AI trading copilot: agents that research markets, generate trading strategies, backtest them and iterate on them together with the trader in a chat. We are building it for serious retail traders who want institutional-grade research tools that until now were available mainly to hedge funds and prop desks.

You will join the Applied AI team, a small product team focused entirely on the engine behind Quant Pilot's agent: AI research, strategy generation and the agent platform that runs them.

The central problem of this role is making agents produce strategies that are statistically sound, not only plausible-looking, and proving it with evaluation. We judge our work by what traders get out of it, not by features shipped.

Models change every few months, so we invest in the engineering around them and prefer to own that code rather than depend on heavy frameworks.

• Own the agent architecture end to end: orchestration, tool loops, context management, sandboxes and model routing, including the quality, cost and latency trade-offs of running it in production.
• Own the contracts between the engine and the product, including how the frontend renders agent events.
• Build and own the evaluation layer: offline evals, backtest-based quality metrics and regression detection for every prompt or model change.
• Bring statistical rigor to strategy evaluation: overfitting detection, robustness checks, and walk-forward and out-of-sample validation.
• Build the data foundations the agent reasons over, with provenance, validation and a consistent model of market entities, and make the system observable enough to diagnose non-deterministic failures in production.
• Shape what gets built: refine requirements with product, set technical direction through RFCs and ADRs, and mentor engineers moving into LLM systems.

• 5+ years of backend or ML engineering, including at least one year building LLM-based systems in production. You understand agent failures one level below the framework and can fix them in your own code.
• A track record of eval-driven development: you can show a quality metric you kept honest over time and how you used it to improve an agent.
• Strong Python and production fundamentals (services, queues, streaming, observability, Kubernetes). You're comfortable following a bug into Type Script or Go, and you treat pipelines, migrations and guards as real engineering.
• Application of statistical principles on backtesting methodology.
• A product mindset: you judge your work by what the user gets, and you take ambiguous goals to shipped, measured results.
• Working proficiency in English (B2+).

• Quant finance background: backtesting frameworks, market data and strategy research, including the pitfalls of backtesting indicator-based strategies.
• Experience with knowledge graphs, ontologies or semantic layers. RAG experience is a plus.
• Experience with Go.
• You have built software on the receiving end of the MCP protocol, such as MCP servers or tools consumed by agents.
• High-load, low-latency systems in fintech or trading.
• Language model fine-tuning or training experience.

Responsible use of AI-assisted development tools is an expected part of our engineering workflow. Engineers use them in daily work while remaining fully accountable for the correctness, security, maintainability and production impact of everything they ship.

We believe in "you build it, you run it". Owning a service end to end means being there when it matters, so this role includes taking part in the team's on-call rotation for the services you own, acting as first responder for production incidents and following our incident response process.