I built an AIOps platform that auto-heals Docker services as a student

Ayush Walunjkar — Thu, 03 Sep 2026 05:06:44 +0000

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.