Deep-search Multi-agent System
A deep-search system where several AI agents work together to research a question and produce an answer, with each agent running as its own serverless function on AWS.
Request path in production
- requestAPI Gateway
- orchestrationcoordinates the agents
- agentsone Lambda function each
- search and retrievalincluding a vector database
- answercombined from the agents’ work
High-level architecture
- Multi-agent deep search
- Instead of one model call, several specialised agents split the research, search in depth and hand their results on to be combined.
- Agents as Lambda functions
- Each agent runs as its own serverless function: stateless, independently deployable and scaled per request, with API Gateway in front.
- Retrieval with vectors
- A vector database gives the agents semantic search over the material they work with.
- Backend services
- Python and FastAPI services around the agents, deployed on AWS alongside the functions.