Python backend systems · AI infrastructure · test automation
Electrical engineer with eight years of aerospace and defense experience across hardware validation, systems integration, qualification, and failure investigation. I now build Python systems that connect APIs, telemetry, local AI, hardware, and operator workflows to measurable evidence.
Role direction: Python Backend Engineer · AI Application Engineer · Automation / Developer Productivity Engineer
| Area | Engineering focus |
|---|---|
| Backend systems | FastAPI services, typed contracts, SQL persistence, asynchronous work, health checks, and failure handling |
| AI infrastructure | Local vLLM inference, request admission, context budgeting, performance measurement, and rollback controls |
| Telemetry and hardware | Binary protocols, CRC validation, serial communication, device detection, and evidence capture |
| Verification | pytest, static checks, containerized integration tests, CI, explicit pass/fail criteria, and reproducible artifacts |
Five-service Python system that receives a fixed binary telemetry frame, preserves the original bytes, validates the interface contract, coordinates durable processing through PostgreSQL and Redis, and returns operator-readable results.
Verified behavior: 146 automated tests · isolated end-to-end verification · raw-byte preservation · idempotent duplicate handling · sequence checks · CRC rejection
Demonstrates: service boundaries, durable state, API design, binary protocol handling, failure-path testing, and Docker Compose integration.
Fail-closed Python workflow for serial and CAN hardware validation. It detects Linux devices, records exact command/response evidence, and prevents software-only checks from being presented as physical hardware validation.
Demonstrates: PySerial, hardware presence gates, structured evidence artifacts, explicit failure behavior, and the boundary between simulation and observed bench results.
These repositories remain private because they contain workstation-specific operating configuration rather than public portfolio material:
- AIWork control plane — executes bounded agent changes in isolated Git worktrees, verifies them independently, requires human ownership attestation, and rolls back failed applications.
- Local AI runtime — versions the vLLM container configuration, health checks, measured routing experiments, and rollback profiles for a local NVIDIA GPU inference stack.
- Qwen request governor — provides OpenAI-compatible request sanitation, context budgeting, admission control, upstream failure handling, and request audit records.
Python · FastAPI · Pydantic · SQLAlchemy · PostgreSQL · Redis · pytest · Docker Compose · Linux · systemd · vLLM · GitHub Actions · serial/CAN telemetry
Code existence is not proof of system behavior. Claims are limited to behavior that was directly exercised, checked against explicit acceptance criteria, and preserved as repeatable evidence.