I'm an Engineering student from Pune, India. Over the
last several months I built WATCHDOG — an AIOps platform that handles
the full incident response loop automatically.
The problem it solves
Traditional monitoring alerts you when something breaks. Then a human
wakes up, investigates, diagnoses, and fixes it manually — at 3am, under
pressure. WATCHDOG handles that loop automatically.
What it does
Detection — 6 parallel algorithms run on every metric:
- Statistical Z-score
- Isolation Forest (ML)
- LSTM deep detector
- Multivariate correlation
- Seasonal detector
- Drift detector
Prediction — Failure prediction and capacity forecasting before
impact. Tells you which service is trending toward failure and how long
you have.
Root Cause Analysis — Causal graph, blast radius estimation,
dependency mapping, change intelligence.
Auto-Heal — Real Docker restarts via Docker SDK. 1.4 second verified
recovery time. But not blindly — every action passes through:
- Confidence gate (≥65%)
- Severity gate (HIGH/CRITICAL only)
- 30s per-service cooldown
- 10 actions/hour rate limit
- Circuit breaker (trips after 3 failures)
Audit — Every detection, decision, and action written to a
tamper-evident cryptographic chain before execution.
Real numbers from engineering testing
- 40/40 reliability tests passing (97%)
- 108 events/sec load tested, 0 failures
- 1.4–1.7s verified auto-heal recovery
- 17/17 RBAC tests
- 18/18 multi-tenant isolation tests
- 10/10 WAL recovery tests
Tech stack
Python, FastAPI, PostgreSQL, Redis (WAL), Docker SDK, JWT auth,
SQLAlchemy, asyncio — 104 REST endpoints, 16 async pipeline workers.
What I'm looking for
I'm not selling anything. I'm looking for DevOps engineers and SREs
who can tell me:
- Does this solve a real problem you face?
- What's missing?
- Would you run this in your environment?
Happy to give anyone a free pilot.
Top comments (0)