Four building blocks for AI in government.
Oddblocks are the strategic product efforts of Oddball Labs, each aimed at a hard problem in government technology. We build them as working prototypes and reusable capabilities, prove them in our own hands, and our delivery teams shape them into finished products for clients.
01 FrontierAgentic Experience
AI-enhanced products and workflows: systems that plan, invoke tools, and execute multi-step tasks rather than produce single-turn responses. Frontier is where agents earn a place in government services, with a human in the loop and standards compliance built in from the start.
- Agents that speak interface. Chat-based agents whose output renders as accessible, standards-compliant interface components, so an answer arrives as a usable government UI rather than a wall of text.
- Workflow agents with checkpoints. Agents that execute a defined sequence of tool calls and decision points for a specific service, with human-in-the-loop checkpoints wherever judgment belongs to a person.
- Agent orchestration on open standards. Model Context Protocol servers that expose tools, data sources, and context to agents in a consistent, discoverable interface, so agentic capability scales to an enterprise instead of living in one app.
- The paperwork, handled. AI-assisted tools that cut the manual burden of compliance documentation, from ATO packages and system security plans to automated extraction, summarization, and form completion.
Featured products
- A11y Pilot. An agent that scans in-flight products against Section 508 and WCAG accessibility standards, so compliance is checked continuously instead of discovered late.
- USWDS-render. Rendering tools that translate agent output into accessible, standards-compliant interface components built on the U.S. Web Design System.
02 ForgeOntology and Data
Products that help the Government define, organize, and govern what its domain knowledge actually means, formalized as ontologies, taxonomies, and knowledge graphs. Forge surfaces the insights and connections that were previously infeasible to find in unstructured or voluminous data.
- Legacy systems, explained. Modernization tools that uncover what a legacy system actually does, recovering the business rules and logic implicit in old code through static analysis and LLM-assisted reverse engineering, and paving a defensible path to rebuild.
- Organizational knowledge, connected. Ontology tools that draw on knowledge across structured and unstructured formats, from documents and databases to transcripts, using retrieval-augmented generation and entity and relationship extraction to surface patterns and gaps, grounded in the client’s own data and mission.
- Data platforms that clean as they go. AI-enhanced platforms that accelerate delivery in ecosystems where cleansing, deduplication, and normalization are prerequisites to everything else.
Featured products
- Business Rules Extraction. Recovers the business rules and logic implicit in legacy code through static analysis and LLM-assisted reverse engineering, paving a defensible path to modernization.
- DONUT. An AI-enhanced data platform that accelerates delivery where data cleansing, deduplication, and normalization are prerequisite objectives.
- Research Findings Explorer. An ontology tool that draws on organizational knowledge across documents, databases, and transcripts, using retrieval-augmented generation and entity and relationship extraction to surface patterns, connections, and gaps.
03 StewardAI and Data Governance
The compliance and accountability layer for AI implementation: the policies, controls, and oversight mechanisms that let agencies develop and deploy AI responsibly at enterprise scale. Adoption without governance is how AI programs stall; Steward is how they keep moving.
- Portfolio oversight. Operations and maintenance tools that help agencies sustain large portfolios of existing applications, custom-coded or low-code, by scanning code, reviewing documentation, and checking availability in real time.
- Standards enforced automatically. Products that scan in-flight work against government standards, starting with Section 508 and WCAG accessibility requirements, so compliance is checked continuously instead of discovered late.
- Governance for agent infrastructure. Enterprise-level controls for agent orchestration: shared authorization, immutable audit logging, attribute-based access control, and fine-grained permissioning across multiplexed services.
Featured products
- MCP Multiplexer. Enterprise MCP orchestration governance: shared authorization, immutable audit logging, attribute-based access control, and fine-grained permissioning across multiplexed MCP servers.
- Medicare MCP. An MCP server exposing Medicare domain tools, data sources, and context to agents in a consistent, discoverable interface.
04 SentinelAI in Cybersecurity
A suite of agentic security products that bring the benefits of AI tooling to established cybersecurity practice, securing agentic interactions and scoring resilience and readiness.
- Guardrails for agents. Input and output filtering, policy enforcement, prompt-injection detection, and AI assurance and red-team testing that support responsible AI in government services.
- Monitoring that understands agents. Security monitoring extended to agentic workflows: detecting anomalous tool usage, data exfiltration, jailbreak attempts, lateral movement, and privilege escalation.
- Zero Trust, scored. Zero Trust Architecture designs, tooling, and scorecards aligned to CISA’s Zero Trust Maturity Model and OMB M-22-09, supporting the federal policies agencies are mandated to implement.
Featured products
- Fenceline. Agentic guardrails: input and output filtering, policy enforcement, prompt-injection detection, and AI assurance and red-team testing for responsible AI in government services.