AI application · case study
CloseFuture AI Assistant.
A retrieval-driven assistant with source citations, specialized agents, and MCP tool integrations.
With a source.Original architecture illustration · source-based
01 / The context
A problem worth solving.
A useful assistant must connect answers to evidence and coordinate actions without repeating side effects when a conversation resumes or a request is retried.
02 / My contribution
The engineering underneath.
Implemented retrieval with citations, intent routing, and specialized search, scheduling, and lead agents. Local stdio MCP servers connect tools, while session restoration and idempotency mechanisms guard against duplicate email dispatches.
Implemented features
- Grounded retrieval and a no-answer fallback
- Search, scheduling, and lead routing
- Calendar and email tool integrations
- Session restoration and sanitized traces
Inside the system
Follow the thinking.
Explore a source-based interaction study. No live data, AI calls, or transactions.
Understand intent
The case study routes a conversation toward specialized search, scheduling, or lead agents.
Illustration of documented behavior
Read the complete flow as text
- Understand intent · ROUTING
The case study routes a conversation toward specialized search, scheduling, or lead agents.
- Ground the answer · RETRIEVAL
Retrieval uses PostgreSQL and pgvector. Responses can cite sources or use a no-answer fallback when context is insufficient.
- Coordinate a tool · TOOLS
Local stdio MCP tools connect calendar and lead workflows. Session restoration and idempotency guard against repeated side effects.
04 / Scope & boundaries
An honest account of the scope.
An independent case-study implementation, without implied employment by or endorsement from CloseFuture. External integrations require their own configuration; this portfolio does not call or depend on them.
Read the implementation.
The repository contains the source and supporting context for this project.
View on GitHub (opens in new tab)