Python Backend Engineer — building reliable, observable systems with async Python, PostgreSQL, and production-grade deployment practices.

📄 View my CV · panpanageas@gmail.com · LinkedIn

Python backend engineer focused on systems that stay predictable under load and failure. Work spans real-time data pipelines, a commercially sold desktop automation platform, open-source developer tooling, and full-stack SaaS — with an emphasis on reliability, observability, and maintainability. Core work is in async Python (FastAPI, SQLAlchemy, asyncio) backed by PostgreSQL, with TypeScript for SaaS and tooling projects.

Three systems I designed, built, and operate end to end. Each is public, deployed or distributed, and documented with its own architecture decisions and known limitations.

Hutsix — commercial Windows desktop automation platform, sold through a public storefront since 2023 (€29.99, 14 releases). A profile-based workflow engine with a trigger evaluator (global hotkeys, pixel, template matching, OCR, YOLOX models), a watchdog scheduler enforcing per-binding cooldowns, and a deterministic input executor behind a focus guard. Ships as an MSI with a SHA-256-hashed release manifest, license activation, and per-binding diagnostics that explain why any trigger did or did not fire. Python, PySide6, OpenCV, PyTorch, CUDA. Maintained end to end: architecture, packaging, distribution, and customer support.

Conductor — PostgreSQL-backed async task queue for Python teams who don't want to run a broker, published to PyPI as conductor-task-queue (v0.2.0). Atomic claiming via FOR UPDATE SKIP LOCKED, retries with exponential backoff, a dead-letter queue, priority routing, recurring tasks, per-task-type circuit breakers, task dependencies, and a gRPC API for polyglot workers. Delivery is at-least-once with idempotent handlers — a distinction the README documents rather than glosses over.

feedstream — real-time AIS maritime ingestion and query service, live on Fly.io. A WebSocket ingestion worker feeding idempotent PostgreSQL writes (ON CONFLICT dedup keys), Redis caching, and a FastAPI query API. Reconnect logic uses exponential backoff with jitter plus an in-process circuit breaker. Ships five architecture decision records, Prometheus/Grafana dashboards, request-ID tracing, structured JSON logs, and an explicit list of accepted trade-offs. Alembic migrations run in CI before each release, and a separate retention process purges expired rows.

task-manager-api — Production-style FastAPI service with async SQLAlchemy, Alembic migrations, PostgreSQL, JWT authentication, Docker Compose, and 85%+ pytest coverage. Non-blocking handlers, migration-first schema management, and strict per-user ownership isolation.

event-driven-task-engine — Queue-backed concurrency engine in Python with priority scheduling and graceful shutdown, built around a clean separation between dispatch and execution. The worker reliability patterns carry over directly to RabbitMQ and Redis Streams architectures.

project_x_public — Distributed platform for dataset generation and model training. FastAPI job API, async dataset/training workers, Redis + PostgreSQL job backend with automatic fallback, API-key-based multi-tenancy, and S3/MinIO artifact storage with retention cleanup. CI and release-acceptance gates included.

agent-pr-firewall — GitHub App that acts as a merge guardrail for teams using human + AI coding workflows. Verifies signed webhooks, evaluates pull requests against configurable policies (secret pattern detection, protected paths, missing issue references, PR size, draft status), and publishes results as a required check run with a managed PR comment. Released to v0.1.1 with a full release cycle and a deployment hardening guide.

ollama-vscode-agent — Local-first VS Code extension that connects to Ollama for streaming chat, grounded workspace context, reviewed code edits (diff preview with explicit apply/reject), and approved terminal command execution.

django-digital-products — Production-ready Django 4.2 e-commerce platform for selling digital products directly to consumers: product management, cart and checkout, Stripe webhooks, automated file delivery, order management, an admin dashboard, Celery + Redis, and Docker Compose.

• RenewalRadar (private) — B2B SaaS that detects customer renewal risk before churn. Combines account data, renewal dates, usage and support signals, billing status, and transparent health scoring with AI-assisted next-best-action recommendations into an operational dashboard. Stack: Next.js, TypeScript, PostgreSQL, Prisma, Stripe, OpenAI, background jobs.
• Listing Copilot (private) — Real estate SaaS for agents: property management, AI listing draft generation, public one-sheet sharing with lead capture, an inquiry inbox and follow-up queue, seller reports, usage limits, and Stripe billing. Stack: Next.js App Router, TypeScript, Prisma, PostgreSQL, Auth.js, OpenAI, Resend, Playwright.
• AI Pipeline Platform (private; public counterpart project_x_public) — Distributed platform for dataset generation and model training with async workers and API-key-based multi-tenancy. Stack: FastAPI, Python, PostgreSQL, Redis, MinIO/S3, GitHub Actions.
• ReplyKit (private) — Manifest V3 Chrome extension for reusable snippet management with page-aware variables and an on-page picker. No build step, JSON import/export, favorites-first sorting. Stack: TypeScript, Chrome APIs.
• Hutsix (commercial source) (private) — The private source repository behind the commercial Hutsix desktop automation product. Stack: Python, PySide6, OpenCV, PyTorch, CUDA, YOLOX.

• B.Sc. Computer Science
• B.Sc. Applied Electronics

• Participated in Google Bootcamp 2022 as part of a team on project Ecodeorama.
• Currently working on FlameverseAI.

Open to Python / backend engineering positions — remote or EU-based — with end-to-end ownership from architecture through deployment and operations.

📫 Contact: panpanageas@gmail.com · LinkedIn · 📄 CV