Multicloud multi-agent system for complex document generation
- 4 díasper document, was ~2 months
- ~85%of the first draft reusable
- 7+models with routing
- 200 MBand up to 10 files per case
The problem
Producing tenders, proposals and complex technical documentation ran on cycles of around 2 months: multiple review rounds and heavy manual coordination between technical, sales and compliance teams.
And the output came out different depending on who did it.
The architecture
Specialised agents coordinated Agent-to-Agent across several cloud providers, routing between 7 or more models by cost, quality and availability.
It takes up to 10 files and 200 MB per case, and covers document ingestion, source prioritisation, contextual retrieval and specialised generation, with concurrency and streaming in the operational interface.
Impact
- From cycles of around 2 months down to 4 days, subject to the internal review process
- First draft reusable at around 85%
- Routing across 7 or more models by cost, quality and availability
- Compliance controls over a set of 101 tests
Stack
- Multi-Agent
- A2A
- Multicloud
- Azure AI Foundry
- Python
- Streaming