About the role
As an ML Engineer at Basis, you’ll own end-to-end projects that bring intelligence into production. You’ll act as the responsible party for systems that help our agents reason, plan, and evaluate themselves — meaning you’ll scope, build, and deliver from first principles. You’ll have full autonomy: plan your projects, define success, run experiments, and decide when your system is ready to ship.
You’ll move fast, instrument deeply, and design for clarity — building the scaffolding that lets models act safely and improve continuously. This is a role for engineers who want to operate like researchers and builders at once: reasoning, experimenting, and shipping systems that get smarter over time.
What you’ll be doing:
- Build and evolve our agent systems
- Design and iterate multi-agent architectures that automate real accounting workflows.
- Encode autonomy boundaries, tool usage, and fallback behaviors that make agents safe and reliable.
- Manage context and memory for coherence across steps; plan and execute agent loops with measurable success criteria.
- Route, evaluate, and optimize models under real-world constraints (latency, cost, accuracy).
- Design evaluation and experimentation frameworks
- Build scalable evaluation pipelines (offline + online) that run hundreds of experiments automatically.
- Define golden tasks, labeling strategies, and metrics that make performance measurable and comparable.
- Instrument the stack to detect regressions, track error taxonomies, and drive closed-loop improvement.
- Use data and experiments to drive product and architectural decisions—not just intuition.
- Engineer for context and retrieval
- Architect prompt stacks and instruction hierarchies that structure model reasoning.
- Build retrieval and indexing pipelines that surface relevant context efficiently.
