Skip to content

← Selected work/02

AI Log Observability Platform

Real-time log ingestion, metrics aggregation and AI-driven remediation — end to end.

Period
May–Jun 2026
Role
Design & implementation
Commits
24
Key metric
Lock-free aggregation · SHA256 dedup · auto-patching
Links
Repo ↗
System diagram
signalr dashboardslive stream01ingesthttp intake02dedupsha256 content hash03aggregatelock-free metrics04detectanomaly rules05ai patchproposed fixredis pipelines · postgres persistence · dockerized e2e — anomalies in, candidate patches out

Problem

Logs are cheap to write and expensive to reason about. At volume, the questions that matter — is this new, is it getting worse, what do I patch — drown in throughput. This platform answers them in real time.

Architecture

An ingestion pipeline backed by Redis handles deduplication (SHA256 content hashing) before anything hits storage. Metrics are aggregated lock-free using concurrent collections — no locks on the aggregation hot path. SignalR streams everything to live dashboards, and an AI diagnostics layer turns anomalies into candidate patches automatically.

Key decisions

  • SHA256 deduplication at ingestion — identical events cost nothing twice.
  • Lock-free metrics aggregation via concurrent collections — correctness without contention.
  • AI-driven diagnostics with automated patch generation — the system doesn’t just flag an anomaly, it proposes the fix.
  • Dockerized end to end — the whole platform, from ingestion to dashboard, comes up with one command.

What it proves

Backend depth on .NET: concurrent data structures, Redis pipeline design, real-time transport with SignalR, and integrating an AI layer as a reliable system component rather than a demo feature.