AdvisorDesk: AI-Powered Advisory Content Platform
A content hub for financial advisory firms, with a CMS, a client app, a RAG assistant that cites its sources, and an agentic MCP content layer.
- Role
- Full Stack Engineer
- Timeline
- 2026
- Stack
- ReactNext.jsTypeScriptFastAPISupabase (Postgres + pgvector)OpenAIMCPDockerAWS
Problem
Financial advisory firms produce a steady stream of written guidance, but their clients cannot easily get answers from it. The guidance lives in documents and portals, so a client with a question has to search through them or wait for an advisor to reply. A generic chatbot does not solve this either. In a regulated space, an answer that cannot be traced back to the firm's own published guidance creates risk instead of value.
Approach
I treated the firm's published content as the single source of truth and built the platform around it.
- A CMS lets advisors create, edit, tag, and publish guidance. Publishing is what makes content available to the assistant.
- A client app includes a RAG assistant that answers questions using vector search over published content only. Every answer is grounded in that content and includes citations that point back to the source guidance.
- An agentic content layer exposes the CMS operations for drafting, tagging, searching, and publishing as MCP tools, so a content manager can run the editorial workflow through natural language instead of manual UI steps.
Architecture & Tech
Two Next.js frontends, the internal CMS and the client app, share one FastAPI backend with Google OAuth. Content and embeddings live in Supabase, which provides Postgres with pgvector. Retrieval runs vector search over published content, and OpenAI generates the cited answers. An MCP server exposes the CMS operations as tools for agentic use. The whole stack is containerized with Docker and deployed on AWS.
Results & Impact
- Clients get instant, cited answers grounded in exactly what the firm has published, and every claim is traceable to a source.
- Advisors manage the content lifecycle in one place, and the MCP layer turns multi-step editorial tasks into single natural language requests.
- The platform covers the full stack end to end, including typed React and Next.js frontends, a Python API, vector search, LLM integration, and cloud deployment.
What I'd do next
I would add an evaluation harness that scores how well answers are grounded in their cited sources, stream assistant responses for a faster feel, and extend the MCP toolset to cover analytics, such as reporting which client questions have no published guidance yet.