back home
07 — selected builds

projects

systems, infrastructure, and applied ml — engineered for performance, resilience, and real-time intelligence.

  1. 01

    private watch parties with sub-second playback drift over a tailscale mesh — no cloud, no port forwarding, no telemetry. jitsi powers voice/video and live reactions across a self-hosted jellyfin stack.

    reactkotlinswiftnode.jsjellyfintailscaledocker
  2. 02

    a 30 fps zero-copy pipeline using yolov8 + bytetrack with 40% memory reduction via cached clahe. ray-casting geofencing and fsm hysteresis eliminate false alarms entirely.

    pythonyolov8bytetrackopencvedge ai
  3. 03

    a scalable go edge service with ip-based token-bucket rate limiting and fingerprint-based bot detection. redis-backed counters, a react admin dashboard, and prometheus monitoring, containerized via docker compose.

    goredisreactprometheusdocker
  4. 04

    a forensic classifier combining mobilenetv3 with error level analysis, trained on synthetic forgeries via poisson blending. sub-550 ms latency through automatic cuda/mps hardware fallback and telea inpainting for ocr.

    pytorchopencvmobilenetv3cudafastapi
  5. 05

    a cbir engine using resnet-50 with gem pooling and two-stage verification via faiss + orb/ransac. color-histogram pre-filtering drives 39 qps throughput, served over websockets.

    pytorchfaissresnet-50websocketsnumpy
  6. 06

    a go agent transmitting metrics securely over grpc to a central server, with time-series storage in postgresql and a react dashboard for live updates and multi-node alerting.

    gogrpcpostgresqlreactdocker
  7. 07

    a secure secrets backend with per-tenant aes-256 isolation and jwt + rbac auth. vault-api integration handles automated rotation with audit logging, surfaced through a react dashboard and deployed with end-to-end encryption.

    gopostgresqlreacthashicorp vaultdocker