We Didn’t Just Automate PRs — We Built a 3-Stage Triage Pipeline That Handles 500+ Repos

Let’s be honest. If you maintain more than a handful of open-source repos, you know the pain. PRs pile up. Issues get buried. Contributors get frustrated and leave.

We were there. Managing nearly 500 repos for a client — a large open-source foundation — with a distributed team in Ho Chi Minh City and Can Tho. The manual triage was killing us. A single senior dev was spending 4+ hours a day just sorting through incoming PRs.

Why Smart Tech Leaders Hire Vietnamese Developers in 2025 (And Why You Should Too)

TL;DR: Vietnam is now a leading destination for offshore software development. With a deep STEM talent pool, stable… ...

Read more

So we stopped reacting. We built a 3-stage automated triage pipeline.

It works. Here’s exactly how we did it.

How AI is Reshaping the Software Development Lifecycle (And Why You Should Care)

TL;DR: AI coding tools are transforming the quy trình phát triển phần mềm bằng AI, cutting development time… ...

Read more

Stage 1: The Webhook Listener (Don’t Poll, Listen)

The first mistake most teams make? Polling the GitHub API every 5 minutes. That’s wasteful. You’re hammering the API for no reason.

Instead, we set up a lightweight FastAPI webhook receiver. Every `pull_request` event hits our endpoint instantly.

python
# webhook_receiver.py
from fastapi import FastAPI, Request, HTTPException
import hmac, hashlib

app = FastAPI()

SECRET = os.environ["GITHUB_WEBHOOK_SECRET"]

@app.post("/webhook")
async def handle_webhook(request: Request):
    body = await request.body()
    signature = request.headers.get("x-hub-signature-256", "")
    
    # Verify the signature
    expected = "sha256=" + hmac.new(
        SECRET.encode(), body, hashlib.sha256
    ).hexdigest()
    
    if not hmac.compare_digest(signature, expected):
        raise HTTPException(403, "Invalid signature")
    
    payload = await request.json()
    if payload.get("action") in ["opened", "synchronize"]:
        pr_data = extract_pr_data(payload)
        # Push to Redis queue for processing
        redis_client.lpush("pr_queue", json.dumps(pr_data))
    
    return {"status": "ok"}

Why this matters: We reduced API call volume by 94% compared to polling. The webhook fires the moment a PR is opened — no delays.

Stage 2: The Classification Engine (Don’t Guess, Analyze)

This is where the real work happens. Our Vietnamese team built a classification engine that runs as a GitHub Action. It analyzes each PR across 3 dimensions:

1. Code diff size — Is this a typo fix or a 5000-line refactor?
2. File types modified — Are they touching core logic or just docs?
3. Changed code patterns — Do they match known bug patterns?

The engine uses a lightweight Python script with `git diff` parsing. No AI model needed for the basic stuff.

Related reading: Why Vietnam outsourcing is Beating India at Its Own Game in 2025

Related reading: Outsourcing Software Development Without the Headaches: A CTO’s Playbook for 2025