AGENTIC AI / ENTERPRISE PLATFORM
Enterprise AI Research Platform
Designing a grounded, multi-step AI research experience that brings internal knowledge and external information into one workflow.
The challenge
Marketing research workflows required substantial manual effort and information gathering across disconnected internal and external sources. The platform was designed around a research process described as approximately 40 hours of manual work.
My contribution
- Designed agent workflows for collecting evidence, synthesizing grounded findings, human review, document packaging and citation checks.
- Connected enterprise search and external systems using retrieval-augmented generation, APIs and Model Context Protocol (MCP) tool integrations.
- Defined evaluation approaches using ground-truth sets, LLM-as-a-judge scoring, groundedness and citation accuracy checks.
- Guided a seven-person engineering team through design reviews, integration testing, user acceptance testing and release readiness.
Architecture & approach
High-level conceptual view of the workflow:
Conceptual architecture simplified for public viewing; not an internal system diagram.
Design considerations
- Human-in-the-loop review before packaging research outputs.
- Observability for latency, tool calls, token usage, cost and workflow outcomes.
- Retrieval and citation validation to support traceability rather than relying on ungrounded model responses.
Outcome & impact
Delivered an enterprise AI platform to production with approximately 26 pilot users, with broader adoption planned. The 40-hour figure describes the manual research process being addressed, not a verified amount of time saved.