๐๐๐ถ๐น๐ฑ๐ถ๐ป๐ด ๐๐ฒ๐๐๐ผ๐ฐ๐๐: ๐๐๐๐ผ๐ป๐ผ๐บ๐ผ๐๐ ๐ช๐ฒ๐ฏ๐ต๐ผ๐ผ๐ธ๐ ๐ฎ๐ป๐ฑ ๐ฅ๐ฒ๐ฑ๐ถ๐ ๐๐ฎ๐ฐ๐ต๐ถ๐ป๐ด
I had university exams last week.
I almost stopped working on this project. Then I realized a lesson. If you cannot build while life happens, you will never finish anything.
I coded at night after my exams. This post is the result.
DevDocAI was working, but it was manual. You had to trigger the pipeline yourself. That is just a script. Real autonomy requires automation.
Phase 5 adds two things:
- GitHub Webhooks: The system listens for PR merges to trigger the pipeline.
- Redis Caching: The system stores status and repo data for speed.
The new workflow:
- A developer merges a PR.
- GitHub sends a webhook to DevDocAI.
- DevDocAI verifies the request.
- The agent pipeline starts automatically.
- Documentation updates itself.
Security is vital. Anyone can send a fake request to a URL. I use HMAC-SHA256 signatures to prevent this. We check the X-Hub-Signature-256 header against a shared secret. If it does not match, we reject the request.
I also added Redis. Querying a database every time a user opens a dashboard is slow. I use three cache keys:
- Repo documentation
- Pipeline status
- User repositories
When a PR triggers an update, I delete the old cache immediately. This prevents stale data.
Technical lessons from this phase:
- Always verify webhook signatures.
- Invalidate cache when data changes.
- Use async tasks for background pipelines.
- On Windows, use psycopg[binary] for LangGraph checkpointers.
Building under pressure taught me that constraints force focus. When you have little time, you build what matters. Do not choose between studying and building. Do both.
Next is Phase 6: The Next.js frontend. This will include a dashboard, a review panel, and a chat interface.
Optional learning community: https://t.me/GyaanSetuAi