AI Engineer

PolyPhaze is a data intelligence platform that makes enterprise data AI-ready by unifying and governing information across systems. The AI Engineer will build production-grade AI agents on the Knowledge Fabric, implement governance and semantic data integrations, deliver client solutions, and explain the resulting architectures and trust mechanisms to technical and non-technical stakeholders.

Responsibilities:

• Design, implement, and deploy AI agents using MCP-compliant infrastructure, defining agent purpose, authority levels, MCP connection scope, and behavioral constraints as code
• Implement the governance envelope for every agent you build; identity and provenance, authorization scope, authority level enforcement, audit hook, and behavioral constraints, baked into the architecture, not stated in a system prompt
• Use multi-model routing to match tasks to the appropriate model tier, open-source, small language model, domain-specific, or frontier, and build cost governance logic that makes consumption visible before it becomes a problem
• Use coding AI tools (Claude Code, ChatGPT Codex, Cursor, or equivalent) to accelerate agent development and use a second model to QA the implementation and generate end-user documentation from the same specification. Own the output
• Connect source systems to the Knowledge Fabric using streaming, polled, or federated connection modes, selecting the appropriate mode for each system based on data residency, regulatory, and latency requirements
• Configure entity resolution, semantic layer definitions, Trust Score thresholds, and freshness gates for the agents and use cases you are delivering
• Implement and validate write-side governance: verification layer, lineage capture, agent identity attribution, and Trust Score propagation on every agent write
• Debug Knowledge Fabric failures, distinguish entity resolution errors, semantic inconsistency failures, freshness gate violations, and write-side corruption from model or orchestration failures
• Work directly with clients to translate business problems into agent architectures, Knowledge Fabric configurations, and deployment plans
• Deliver working, tested, governed agents, not prototypes, within realistic project timelines
• Explain what you built to technical and non-technical stakeholders: what the agent does, what authority it has, how it is governed, how its outputs can be trusted, and how to interpret the Trust Score on every output
• Participate in client-facing AI readiness assessments, identifying Knowledge Fabric gaps, authority level misconfigurations, and missing governance infrastructure
• Contribute to PolyPhaze's engineering practices: agent design patterns, governance templates, domain ontology extensions, and reusable MCP connection configurations
• Stay current on MCP & A2A adoption, Knowledge Fabric capabilities, frontier & open-source model quality, and the enterprise AI stack, and bring that currency into client engagements, not just internal conversations

Requirements:

• Proficiency in Python and at least one statically typed language (TypeScript, Go, or Java). You write clean, tested, and maintainable code
• Hands-on experience building and deploying AI agents or LLM-based systems in production, not just in notebooks or demos, in production environments where failures have consequences
• Working knowledge of MCP: what it is, how MCP servers are implemented, what an Enterprise MCP Gateway does, and where MCP ends and the semantic layer begins
• Experience with at least one orchestration framework (LangGraph, AutoGen, CrewAI, or equivalent). Understanding of what A2A adds and why the interim approaches accumulate technical debt
• Familiarity with multi-model deployment: when to use frontier models, when to route to smaller or open-source models, and how to implement routing logic that makes cost visible and governable
• Understanding of semantic data architecture: entity resolution, referential integrity, semantic consistency across domain boundaries, and why these properties matter for agentic AI specifically
• Ability to read and configure knowledge graph structures, semantic layer definitions, and Trust Score computation logic, not just consume them
• Experience with at least one data platform that the Knowledge Fabric reads from: Snowflake, Databricks, BigQuery, or equivalent. Understanding of streaming, polled, and federated access patterns
• Appreciation for why write-side governance matters and what governed writes require even if you have not yet implemented a full verification layer
• Ability to distinguish the layer where a failure originated, enterprise application, data platform, Knowledge Fabric, MCP layer, orchestration, or model, and direct remediation effort at the right place
• Comfort with the governance envelope as an engineering requirement, not a compliance box: authority levels enforced architecturally, audit hooks live before the agent moves to its next step, behavioral constraints tested before production promotion
• Honest about what you do not know and fast at closing the gap. This work spans a stack that changes every six months and requires people who update their mental models when the evidence requires it
• Client engagements involve occasional travel
• Direct experience implementing a Knowledge Fabric or comparable semantic and trust layer above an existing enterprise data estate
• Experience with ABAC (Attribute-Based Access Control) for agent populations, not just RBAC, and with designing authority level frameworks that hold up as agent populations scale
• Production experience with A2A: cross-vendor agent coordination, Agent Card design, maximum authority ceiling enforcement for external agents
• Domain depth in one or more verticals where the Knowledge Fabric provides pre-built ontologies: financial services, insurance, healthcare, manufacturing, defense
• Experience designing AI observability infrastructure: decision tracing, ROI observability, and operational AI monitoring that catches drift before it produces consequences
• Background in enterprise data architecture, data mesh, data products, or data warehouse design, with direct experience of the gap between data mesh governance and what agentic AI actually requires