Job Description
Staff AI Engineer — AI/Crypto Fintech | Seed Stage | Backed by Tier-1 Investors
Location: Americas preferred (US > Canada/LatAm > Europe > Dubai > India) | Remote Compensation: $175,000–$250,000 USD base + 1% equity + team bonuses + pro-rata 2026 token launch participation Stage: Seed | 11–50 employees
About the Company
Our client is an AI-native fintech startup building autonomous trading technology. Backed by leading crypto-focused VCs, they have generated significant trading volume with zero paid acquisition and maintain strong user retention metrics. The founding team has deep crypto and onchain product experience, with a prior unicorn venture. Recent senior hires come from top names in the crypto infrastructure space.
The platform is a purpose-built, event-driven execution system for autonomous trading agents operating with real capital. Infrastructure includes a plugin runtime, rules engine, MCP execution layer, and a high-throughput data pipeline (Redis, Postgres, ClickHouse).
About the Role
This is a Staff-level, individual contributor role — hands on keyboard, writing Python, shipping to production. You will own the intelligence layer: building the system where a fleet of autonomous trading agents learns from itself and improves continuously.
The role is approximately 70% reinforcement learning and the full learning loop, and 30% model hosting, inference infrastructure, and LLM tooling. The runtime and data infrastructure already exist — you are building on top of a working platform with real, measurable outcomes on every trade.
Key Responsibilities
- Design and implement the feedback loop connecting trade outcomes back to strategy improvement — signal selection, risk parameters, position sizing, and timing
- Build the evaluation framework that identifies which signals, market conditions, and agent configurations predict profitable outcomes versus noise
- Develop automated strategy generation, backtesting, and deployment pipelines
- Detect market condition shifts and adapt fleet behavior accordingly
- Build performance attribution that decomposes trade outcomes into component drivers and feeds insights back into strategy design
- Manage fleet coordination — concentration risk, capital allocation, and exploration vs. exploitation balance
- Build the telemetry and data capture layer that makes continuous learning possible
- Evaluate and implement the right model hosting strategy (build vs. buy decision) — currently using external LLM APIs (Anthropic, Google); no internal inference stack exists yet
- Optimize inference for many concurrent agents with structured decision outputs at scale
Requirements
- Closed-loop production system (non-negotiable): You have built a system where a model makes a decision, that decision executes in the real world, the outcome is measured, and that outcome automatically feeds back into improving the next decision — live and continuously, not batch retraining or manual incorporation
- RL / online learning experience: Practical understanding of learning from real-world outcomes rather than static datasets; specific technique matters less than the mindset and production experience
- Full-stack ML ownership: You build data pipelines, deploy models, and measure outcomes end-to-end; Python primary, comfortable with Go or TypeScript for production services
- High-stakes domain experience: Finance is a strong preference but not required — robotics, autonomous vehicles, game AI, ad bidding, supply chain, recommendation systems all count; the loop structure is what matters
Bonus
- LLM fine-tuning and serving (PEFT/LoRA, vLLM, TGI)
- Multi-agent system design
- Financial ML — signal generation, alpha research, portfolio optimization
- Onchain/DeFi experience
What Will Get You Rejected
- Never deployed ML in production
- Closed-loop that relied on manual incorporation of findings into the next model version — that is monitoring, not a loop
- Prompt engineering focus rather than agents that learn from experience
- No hands-on experience deploying AI/open-source models
Interview Process
Fast — target first call to offer within 2 weeks.
- Intro call with founders (60 min) — fit, motivation, closed-loop experience
- Technical deep-dive (60 min) — system design: "Design the learning system for a fleet of 50 autonomous trading agents"
- Possible paid trial project (1 week, part-time) — real problem from the stack, compensated
