Enterprise AI Copilot

Enterprise retail teams running AI-powered shelf monitoring had 17 separate intelligence domains (pricing, promotions, assortment, compliance, forecasting, etc.) but no unified way to query across them. Analysts bounced between dashboards and spreadsheets to answer cross-domain questions. Leadership wanted a natural language interface but existing tools couldn’t handle the domain complexity.

Mapped the full 17-agent domain architecture by studying the existing AI platform’s capabilities and gaps. Identified that the core challenge wasn’t building another chatbot—it was routing questions to the right domain agent with the right data context. Built a prototype query router and tested it against 14 real business questions to validate coverage.

Designed a multi-agent copilot with intelligent query routing: each question is classified to 1 of 17 domain agents, which pulls structured evidence from PostgreSQL via a data layer, generates contextual responses with citations, and produces actionable recommendations. Mock mode enables demos without live database access. Artifact generator creates formatted reports for executive review. Deployed on Railway with health monitoring via Grafana dashboards.

• Agent routing over single-model RAG — domain-specific agents with tailored system prompts produce dramatically better answers than generic retrieval
• Mock mode as a first-class feature — enabled executive demos weeks before live data integration was complete
• FastAPI over Flask — needed async endpoints for concurrent agent calls and streaming responses
• Railway deployment over internal hosting — gave stakeholders a URL they could share without VPN or setup

• 17-agent copilot covering pricing, promotions, assortment, compliance, forecasting, and 12 more retail domains
• Demo’d to VP Product and company president—validated the concept for further investment
• 13,000+ lines of production Python with full data layer, agent registry, and governance spec
• Mock mode handles 14+ question categories for demos without live database dependency
• Deployed on Railway with health monitoring, Grafana integration, and Perplexity AI search