Donovan
Wright
class Developer: def __init__(self): self.name = "Donovan Wright" self.based_in = "Wisconsin, USA" self.languages = ["Python", "JavaScript", "Bash", "SQL"] self.interests = ["AI/LLMs", "homelab", "embedded hw"] def say_hello(self): return f"Hey — I'm {self.name}. Let's build something." me = Developer() print(me.say_hello())
// Technical Skills
Languages
Infrastructure
Databases
AI & Data
Frameworks
Core Competencies
// Experience
- Built Python automation to validate large datasets and QC AI training inputs — cut manual review time by 30%.
- Applied RLHF methodologies to evaluate AI-generated code, reducing hallucinations and improving model logic.
- Collaborated with data architects to analyze complex datasets and refine ML algorithms.
- Assessed search and advertising relevance pipelines to ensure high-fidelity training data.
- Architected and maintained web applications using modern JavaScript frameworks and server-side scripting.
- Designed relational database schemas and optimized SQL queries for performance at scale.
- Managed full SDLC from requirements through deployment and security auditing.
- Served as technical liaison translating business requirements into scalable software solutions.
- Maintained application integrity across lab and administrative networks.
- Implemented monitoring protocols that measurably reduced system downtime.
// Projects
A professional-grade open-source smart thermostat on Raspberry Pi with a FastAPI backend and React dashboard. Includes "Eco Intelligence" — weather-aware adaptive hysteresis, predictive pre-heat/pre-cool, real-time heat-loss analysis, BME680 IAQ monitoring, and multi-room remote sensor averaging. Features IP-based read/write access control, energy cost estimation, filter health tracking, and a custom 3D-printable enclosure.
A lightweight Python daemon that continuously pings all LAN devices, tracks latency history in SQLite, and fires webhook alerts via a self-hosted ntfy instance when a device drops or latency spikes past a configurable threshold. Ships with a small React/FastAPI dashboard showing per-device uptime heatmaps.
A fully self-hosted homelab running Proxmox VE and Unraid as hypervisors, with Nginx Proxy Manager handling reverse proxy, SSL termination via Let's Encrypt, and internal DNS. All services are containerized via Docker for isolation and zero-downtime rolling updates.
A configurable Python test harness for batch-evaluating LLM output quality. Runs prompt suites against a target model, scores responses on correctness, coherence, and safety, and writes structured reports to PostgreSQL. Supports pluggable rubrics and parallel async execution for high-throughput evaluation pipelines.
A cron-driven Python service that snapshots PostgreSQL and SQLite databases, compresses and AES-256 encrypts archives, and pushes them to Backblaze B2 with retention policy enforcement. Includes a CLI for restore and integrity verification, and push notifications via ntfy on success or failure.
A self-hosted web app for managing RLHF annotation tasks. Presents prompt/response pairs to reviewers with a configurable rubric scoring UI, persists decisions to PostgreSQL, and exports JSONL datasets ready for fine-tuning. Supports multi-annotator sessions with inter-rater agreement stats.