Overview

I led the research, design, development, and testing of an internal GPT Store powered by multiple AI agents, each leveraging Retrieval-Augmented Generation (RAG) and domain-specific knowledge bases. The solution was built to support key enterprise functions, including Product, Legal, Tech Due Diligence, and Incident Management.

Problem

Employees across departments faced challenges accessing fragmented institutional knowledge. Insights were scattered across Confluence, ServiceNow, and Box folders—making it time-consuming to answer even routine questions like “What’s launching next month?” or “Have we done diligence on this vendor?”

Solution

I designed a multi-agent GPT Store tailored to each domain’s needs, each agent backed by a specialized knowledge base and built-in RAG pipelines. These agents could parse structured and unstructured content and return annotated answers with source citations.

Key agents included:

• Product Roadmap Agent – Understands timelines, launches, and product impacts (e.g., “Which products impact high net worth clients in Q4?”)
• Tech Due Diligence Agent – Synthesizes past diligence memos and vendor risk assessments
• Legal Agent – Surfaces past contract clauses and legal playbooks
• Incident Management Agent – Answers questions about historical incidents, remediations, and related change records

Core Capabilities

• Session Management: Rename, pin, and delete sessions to create persistent, organized dialogues
• Feedback Mechanism: Users can rate responses, suggest improvements, and request knowledge base updates
• Source Attribution: Every response is traceable back to its source, building trust and auditability
• Feedback-to-Feature Loop: Internal feedback is routed to the product backlog for rapid iteration
• Complex Query Support: Handles both simple and compound questions, with natural language understanding of product, legal, and incident taxonomies

Outcomes

• Reduced time-to-answer for complex internal queries by 60%
• $360,000 overhead cost-cutting
• Accelerated decision-making in product and compliance reviews
• Enabled non-technical users to self-serve previously siloed insights
• Created a scalable framework for future agent expansion