We build agentic AI systems for institutional investors, powered by two engines: OmniContext™, our hybrid context engine, and SmartOrch™, our agentic orchestration engine.
Building and deploying AI applications
- Multi-agent workflows in LangGraph, LangChain and Google ADK — routing, delegation, durable execution, human-in-the-loop
- Hybrid retrieval: knowledge graph (Neo4j/Cypher) + vector (pgvector, Qdrant) + SQL, with query routing and reranking
- Gemini, OpenAI, Azure OpenAI and Anthropic, with model-agnostic routing and fallback
- Agent harness — tools, MCP, guardrails, structured outputs, context and token budgeting
- Eval infrastructure — golden datasets, regression suites, grounding and hallucination checks
- Production tracing: model, prompt version, retrieved span, tool call, approver
Software engineering fundamentals
- Python (FastAPI, Pydantic, asyncio) and Node.js/TypeScript services; React/Next.js front-ends
- Postgres and Firestore modelling; document ingestion, entity resolution, schema-drift detection
- Docker, Kubernetes, Terraform, CI/CD on GCP, Azure or AWS
- SSO/RBAC, private networking, secrets management, audit logging
- Deployment into client cloud, on-prem and restricted environments — including open-weight serving (vLLM, Ollama)
Orchestrating agents
- Decomposing work into tasks an agent can complete, with the context to make that likely
- Setting up tests and feedback loops for longer unsupervised runs
- Reviewing agent output critically — you own everything that ships under your name
- Building skills, tools and MCP servers so agents are useful on our codebase
Shaping the build
- Scoping ambiguous client problems into something shippable
- Taking a technical position and defending it, with nobody senior to defer to
- Knowing when a workflow doesn't need an agent
You
- 4+ years shipping production software, full stack in Python and TypeScript
- Built a RAG system and then fixed it; can talk about failure modes, chunking, reranking
- Production experience with an agent framework — not tutorials
- You write evals and have caught a regression before a user did
- Strong SQL; graph databases or able to pick them up fast
- Docker, Kubernetes, CI/CD and at least one major cloud
- Comfortable in front of a client, not just a codebase
Bonus: entity resolution · text-to-SQL · MCP/A2A · Vertex AI or Azure OpenAI in production · on-prem or regulated delivery · financial services domain
We're hiring two engineers to expand the core team.