About Thira
Thira is the CIO's trusted agentic AI platform that continuously learns and executes repetitive IT work. Enterprises will spend $6.3 trillion on IT this year, and roughly 40% of it goes to people doing repetitive work across systems that were never built to work together. The work arrives faster than any team can absorb it, the backlog never clears, and there's no capacity left for the work that moves the business forward. Thira learns how an enterprise actually runs, finds a better way to deliver each outcome, and executes the work end to end across every system, with every action governed. The result is measurable ROI and employees whose problems are resolved before they notice them. Thira is backed by a $21M seed round led by Madrona and led by the co-founders of Apptio, and we're building the team that will define this category.
The role
Thira's agents are only as smart as the systems they learn from. As our AI Systems Engineer, you'll own the knowledge graph that models every enterprise IT environment, and the learning loop that turns each human resolution into a reusable rule, making the platform measurably smarter with every ticket. You'll own the stateful, long-lived systems underneath the orchestration layer: the part that makes Thira defensible and that competitors can't copy by swapping an API key.
What you will own
- Design and own the IT knowledge graph, modeling relationships between users, devices, software, permissions, services, and incidents in a queryable, agent-readable form.
- Build and iterate on the Pattern Miner agent: watch human resolutions, generalize reusable rules, and write them back to the knowledge graph with appropriate confidence scoring.
- Own the Runbook Synthesizer: transform observed resolution patterns into structured, executable runbooks that agents can run autonomously going forward.
- Architect the human review gate and active-learning pipeline, deciding when a learned rule is trusted enough to run without human approval.
- Design the feedback schema and preference-data pipeline for eventual fine-tuning of domain-specific models on Thira's resolution data.
- Own evaluation for the learning systems: how do you measure whether a rule is correct, and whether a runbook is safe to automate?
What we are looking for
- 3+ years building production AI or data systems: knowledge graphs, recommendation systems, learning pipelines, or closely related work.
- Hands-on experience with graph data modeling: schema design for real-world domains, not just toy examples.
- Strong understanding of preference learning, feedback loops, or active learning: you've built systems that improve from human signal.
- Familiarity with embedding models, semantic similarity, and retrieval at the knowledge-graph layer (GraphRAG or equivalent).
- Strong in Python and Java, comfortable with async systems, and rigorous about data quality and schema evolution.
- High bar for correctness in autonomous systems: you think carefully about when a system should trust itself vs. escalate.
Strong plusses
- Prior experience building backend systems or full-stack work.
- Ontology design or semantic-web technologies (RDF, OWL, SPARQL) in enterprise data contexts.
- Background in ITSM, IT operations, or enterprise infrastructure: you understand what a CMDB is and why current ones fail.
- Prior work on RLHF, DPO, or fine-tuning pipelines for domain-specific LLM adaptation.
- Familiarity with graph neural networks or graph-based retrieval for reasoning over structured enterprise data.
- Experience building evaluation frameworks where correctness is hard to measure (generalization quality, rule precision/recall).
Location
Bellevue, WA
How to apply
To apply, email us and tell us where you have done this before, why Thira, and a few links. We read every application.