Applied AI-Agent Engineer

Sapience AI Corporation•Seattle, WA

5d•$204,000 - $216,000

About The Position

This role turns AI capability into product behavior members can rely on. You build the applied AI and agentic systems that let MINERVA understand a request, reason over a community’s knowledge, and take useful action. You work where models, retrieval, reasoning, and tools come together: designing agents, grounding them in the KO graph and the COGENT architecture, and making them dependable enough for real communities. You sit between product and the deeper AI stack, and you are the person who makes the platform actually do the intelligent thing, well and safely. Raw model capability is not a product. Members need systems that understand what they are asking, reason over the right knowledge, and act reliably, without confident mistakes. Building agents that are genuinely useful and trustworthy is hard: grounding, tool use, orchestration, and honest handling of uncertainty all have to work together. The Applied AI and Agent Engineer builds that. You turn models and reasoning into applied systems members trust, and you keep them dependable as the platform grows.

Requirements

Five or more years in software engineering, with strong recent work in applied AI or ML.
Hands-on experience building agentic systems, LLM applications, or retrieval-augmented systems in production.
Strong understanding of grounding, tool use, and the limits of models.
Experience making AI behavior safe, reliable, and observable.
Strong Python, plus solid software engineering fundamentals.
Rigor about evaluation and honest handling of uncertainty.
Care for member trust and safety.

Nice To Haves

Experience with agent frameworks, orchestration, and planning.
Familiarity with knowledge graphs and neuro-symbolic reasoning.
Experience with retrieval, embeddings, and hybrid search.
Experience productionizing AI features at scale.
Domain understanding of professional or knowledge-intensive communities.

Responsibilities

Design and build the agents that let MINERVA understand requests, reason, and take action.
Own orchestration, planning, and tool use, and keep it dependable.
Design for the limits of models, not just their strengths.
Ground agent behavior in the KO graph and the COGENT architecture so answers rest on real community knowledge.
Build retrieval that surfaces the right knowledge at the right time.
Reduce confident errors through grounding and verification.
Integrate neural and symbolic reasoning into applied behavior in partnership with neuro-symbolic AI.
Decide where a model should reason and where structure should carry the load.
Turn reasoning advances into product behavior.
Build the tools and actions agents use, including safe access to systems and data.
Make actions reliable, observable, and reversible where they should be.
Connect applied AI to the systems a community already runs.
Build the guardrails that keep agent behavior safe and within bounds.
Handle uncertainty honestly, and make provenance visible.
Treat member trust as a design constraint.
Build evaluation for agent quality, including accuracy, groundedness, and safety.
Instrument applied systems so you can tell whether they are working in production.
Iterate on real behavior, not just offline tests.
Take applied AI from prototype to dependable production, in partnership with ML infrastructure.
Balance capability against latency, cost, and reliability.
Own the behavior members experience, not just the demo.