Featured workflowchangegraph
ChangeGraph
A multi-agent repo analysis tool built as a developer playground at a US company, collaborating with 2 devs and the director — auto-generates docs, architecture diagrams, and changelogs with 150% performance gains from parallel orchestration.
stack
- Python
- FastAPI
- React
- PostgreSQL
- Tailwind
- Docker
- SQLAlchemy
Architecture
- Domain-specialized agents — README, architecture, API reference, security audit, and changelog agents, each with own prompts and output schema
- DAG orchestration — agent dependencies modeled as a directed acyclic graph; independent agents run in parallel batches of 3-4
- Immutable shared context — each agent gets its own copy of repository context to avoid mutation conflicts
- Producer-consumer writes — agent results written back to PostgreSQL via producer-consumer pattern
- Repository parsing pipeline — preprocessing extracts file trees and entry points into a queryable context map
- Incremental analysis — diffs repo state against last run; only affected sections routed to relevant agents
Results
- 150% performance improvement over sequential execution
- 5 documentation types from a single repository analysis
- Incremental mode cuts re-analysis to seconds for small changes
- Scales to 1000+ file repositories without context window overflow
Action · Send message
Hiring for a remote AI engineering role?
Contact me