Distinguished Agent Engineer
fastino.ai24 days ago
Responsibilities
- Design and own architectures for agentic systems built on Fastino models, including planning, tool use, memory, and multi-agent orchestration.
- Evaluate where specialized models outperform frontier LLMs using benchmarks and real customer workloads.
- Build evaluations, telemetry, and feedback loops to measure agent performance and diagnose failures.
- Partner with research to turn new model architectures into agent capabilities and influence future research priorities.
- Improve agent latency, reliability, concurrency, inference serving, and graceful failure handling at scale.
- Drive data and fine-tuning strategies for agent-specific tasks such as tool calling, extraction, and routing.
- Write core production code and set technical direction across teams.
Requirements
- 5+ years of software engineering experience.
- 3+ years building LLM or ML systems in production.
- A track record of shipping agentic systems relied upon by real users.
- Deep experience with prompt and context engineering, tool and API design, vector databases, Kubernetes, and inference serving.
- Strong understanding of model behavior, fine-tuning, evaluations, failure modes, and latency/cost tradeoffs.
- Experience setting technical direction across teams.
- Clear written communication and the ability to simplify complex systems.
- Bonus: experience building high-concurrency microservices for agent workloads.
- Bonus: familiarity with GLiNER or other information extraction architectures.
- Bonus: open-source contributions, papers, or talks on agents or LLM systems.
Benefits
- Full-time position.
- Located in San Francisco or remote.
Tech Stack
Categories
About fastino.ai
Fastino builds task‑optimized small language models and Pioneer, a fine‑tuning and inference system that lets developers deploy and continuously improve open‑source models without code. It targets enterprise AI teams that need accurate, fast, and secure on‑prem or cloud deployments, including near‑instant CPU inference. The privately held company’s research includes the open‑source GLiNER/GLiNER2 models with millions of downloads, and it raised a seed round in 2025.