The challenge
Advisors juggled several tools for routine tasks. They wanted to ask in plain words and have the work done.
What we built
- Text, voice and image requests turned into validated tool calls
- LangGraph orchestration with OpenAI and Gemini Vision
- FastAPI services, PostgreSQL, Redis and Docker deployment
The full story
The situation
Investment advisers juggled several tools for routine tasks. The goal was to ask for something in text, voice or an image and have it done.
What we built
An AI agent platform. Requests come in through FastAPI, LangGraph and LangChain decide which tool to use, OpenAI and Gemini Vision interpret text and images, and every action is checked against business rules before it runs.
How a request flows
- A request arrives by text, voice or media
- The AI layer interprets it
- It becomes a structured tool call
- The call is validated against business rules
- The action runs and a clear answer comes back
Stack
Python, FastAPI, LangGraph, LangChain, OpenAI, Gemini Vision, PostgreSQL, Redis and Docker on a Linux VPS. The design is modular, so new tools and workflows can be added without rebuilding the core.