Staff AI Engineer

Syndesus•Remote,

1d•$175,000 - $250,000•Remote

About The Position

A well-funded seed-stage startup building the next generation of autonomous trading technology. Backed by leading crypto-focused venture capital, the company has driven significant trading volume with zero paid acquisition and strong retention metrics. The founding team are crypto and onchain veterans with a prior unicorn venture. The platform is a purpose-built execution system for AI agents operating with real capital around the clock — the infrastructure, data pipelines, and runtime are already in production. You are building the intelligence layer on top of it.

Requirements

A production closed-loop system — model outputs drove real-world actions, outcomes were measured, and that feedback automatically improved the next decision. Not a batch retrain. Not a dashboard with manual follow-through. A live, wired loop.
Practical RL or online learning experience — you understand the challenges of learning from real-world feedback rather than static datasets
Full-stack ML ownership — you build the pipeline, deploy the model, and own the outcome; Python primary, comfortable with Go or TypeScript in production services
High-stakes sequential decision-making domain experience — finance preferred but not required; robotics, autonomous vehicles, game AI, ad bidding, and supply chain all transfer

Nice To Haves

LLM fine-tuning and open-source model serving in production (vLLM, TGI, PEFT/LoRA)
Multi-agent system design
Financial ML — signal generation, execution optimization, portfolio construction
Onchain or DeFi experience

Responsibilities

Design and implement the pipeline that connects live trade outcomes back to strategy improvement — signal quality, position sizing, timing, risk parameters
Build the evaluation framework that separates genuine predictive signal from noise across agents, market conditions, and configurations
Automate the strategy generation and testing cycle — the system should explore new configurations, validate them against real fleet data, and surface deployment candidates
Detect regime shifts in market conditions and adapt fleet behavior accordingly
Decompose every trade into its component drivers — signal quality, execution efficiency, exit timing — and wire those attributions back into strategy design
Manage fleet-level coordination: concentration risk, capital allocation, and the exploration vs. exploitation balance
Build the telemetry and data capture layer that makes all of the above possible
Own the build-vs-buy decision on model hosting — evaluate proxied external APIs versus fine-tuned models on owned infrastructure and execute the chosen path
Determine whether domain-specific training on trading data meaningfully outperforms prompted general-purpose models — then build the pipeline to act on that answer
Optimize inference for the specific demands of a large autonomous agent fleet: concurrent agents, structured outputs, cost efficiency at scale
Build the agent telemetry layer capturing every decision, signal score, and evaluation across the fleet