Panagiotis Panageas

Python Backend Engineer

Reliability, observability, and systems that stay predictable under load and failure.

panpanageas@gmail.com +30 697 210 7200 github.com/Archangel-77 LinkedIn Messinia, Greece · Remote (EU) Languages: Greek (native), English (professional)

In production
feedstream.fly.dev
On PyPI
conductor-task-queue v0.2.0
Paid product
14 public releases
Design record
5 architecture ADRs

Profile

Backend engineer who builds and operates systems rather than demos. I ship software commercially — a paid Windows automation product with 14 public releases — and run a real-time ingestion service in production. My work concentrates on the failure modes most portfolios skip: idempotent writes, circuit breakers, retries with backoff and jitter, dead-letter queues, and the observability that shortens recovery. I write down the trade-offs I accept instead of hiding them, because an undocumented decision is a future outage.

Flagship Systems

Hutsix — commercial Windows automation platform

Profile-based workflow engine with GPU-accelerated computer vision, distributed as a paid product.

  • Designed a trigger evaluator covering global hotkeys, pixel colour, template matching, OCR, and YOLOX object models, behind a watchdog scheduler that enforces per-binding cooldowns and a focus guard restricting input to a target executable.
  • Built release engineering around a SHA-256-hashed manifest and license activation, and added a preflight gate that fails the build when the configured app version disagrees with the version being published — after diagnosing exactly that mismatch on a clean install.
  • Shipped per-binding diagnostics that explain why a trigger did or did not fire, replacing user guesswork with a traceable execution timeline.

Stack  Python · PySide6 · OpenCV · PyTorch · CUDA · YOLOX · EasyOCR · MSI packaging · license activation

Conductor — PostgreSQL-backed async task queue

Reliable background jobs for Python teams that do not want to operate a message broker.

  • Implemented atomic claiming with FOR UPDATE SKIP LOCKED so a task is never handed to two workers concurrently; delivery is at-least-once, with idempotent handlers named as the contract in the docs.
  • Built retries with exponential backoff, a dead-letter queue with retry and discard, priority routing, recurring cron schedules, per-task-type circuit breakers, and dependency chains that propagate BLOCKED state.
  • Instrumented with Prometheus metrics, health endpoints, and structured logging; added a React dashboard, a gRPC API for polyglot workers, and deployment artifacts for Docker, Kubernetes, and systemd.

Stack  Python 3.11+ · asyncio · PostgreSQL · asyncpg · Pydantic · gRPC · Prometheus · FastAPI + React

feedstream — real-time AIS maritime ingestion and query service

Continuous ingestion of an external event stream, stored durably and served with low-latency queries.

  • Ingestion worker consumes an upstream WebSocket feed and writes idempotently using deterministic dedup keys with ON CONFLICT DO NOTHING, so replayed events and process restarts cannot create duplicate rows.
  • Reconnection uses exponential backoff with jitter plus an in-process circuit breaker; the query API serves cursor-paginated responses behind Redis caching and Redis-backed per-endpoint rate limiting.
  • Instrumented end to end — Prometheus metrics scraped from both the API and worker processes, provisioned Grafana dashboards, request-ID propagation, structured JSON logs, and a separate retention job that purges expired rows.

Stack  FastAPI · SQLAlchemy 2.x async · PostgreSQL 16 · Redis 7 · Alembic · Prometheus · Grafana · Fly.io · GitHub Actions

Engineering Practice

Languages
Python, TypeScript, SQL, Bash, JavaScript
Reliability
Idempotent ingestion, backoff with jitter, circuit breakers, dead-letter queues, graceful shutdown
Backend
FastAPI, SQLAlchemy 2.x (async), Pydantic v2, Alembic, asyncpg, aiohttp, Django / DRF
Data
PostgreSQL 16, Redis 7, Prisma, SQLite; cursor pagination, schema migrations in CI
Architecture
Event-driven design, queue-based workers, state machines, multi-tenant APIs, at-least-once delivery
Observability
Prometheus, Grafana, structured JSON logging, request tracing, health and readiness endpoints
Infrastructure
Docker, Docker Compose, GitHub Actions, Fly.io, Kubernetes manifests, systemd
Quality
pytest, pytest-asyncio, Vitest, Playwright, Ruff, MyPy, pre-commit
Vision / ML
OpenCV, PyTorch, CUDA, YOLOX, ONNX Runtime, EasyOCR
Desktop
PySide6 / Qt

Experience

Independent Software Engineer

Design, build, and commercially distribute software end to end.

  • Own the full lifecycle: architecture, implementation, packaging and release engineering, licensing, documentation, and direct customer support.
  • Operate production services on Fly.io with GitHub Actions CI/CD, secrets held outside the repository, and Alembic migrations applied before each release so the schema is always ahead of the running code.
Hellenic Navy — Submarine Operations

High-consequence operational environment with no tolerance for unverified assumptions.

  • Operated and maintained complex systems under strict procedure, where faults had to be diagnosed correctly the first time and deviations documented rather than improvised around.
  • That habit — check before acting, record what changed, never assume the system is in the state you left it — is the same discipline I apply to release engineering and incident response.

Other Projects

Education