portfolio  email  linkedin  github

I build the retrieval, multimodal input, provider integration, evaluation, and backend reliability layers that turn model capability into useful systems.

• focus RAG, context engineering, multimodal/voice AI, structured outputs, and evals
• backend Python, FastAPI, asyncio, APIs, webhooks, caching, retries, and idempotency
• providers OpenAI, Anthropic, Gemini, DeepSeek, and Azure AI Foundry
• currently building VolyxAI and independently developing Volyx Lens
• status open to AI Engineer and ML Engineer opportunities

Python is my primary engineering language. My JavaScript and native product work is AI-assisted, with architecture, security boundaries, testing, and verification kept explicit.

# AI & ML

Python OpenAI Anthropic Azure AI Foundry Deepgram scikit-learn XGBoost

# Backend

FastAPI asyncio Node.js Express SQLAlchemy SQLite Socket.IO

# Frontend / Apps

React Vite Electron Swift JavaScript

# Delivery / Automation

Docker n8n GitHub Actions Playwright Azure

_{Electron · JavaScript · Swift}

Privacy-oriented macOS assistant for screen, voice, meeting, and coding workflows. Includes provider routing, native capture components, Keychain-backed credentials, and explicit sharing controls. product site →

_{FastAPI · React · Telegram}

Read-only Solana and Ethereum analytics with bounded concurrency, caching, retries, partial-failure reporting, and automated API/browser verification. live demo →

_{Python · XGBoost · FastAPI · Streamlit}

Leakage-resistant chronological evaluation with out-of-fold stacking, calibrated XGBoost and Poisson models, frozen holdouts, and explicit artifact provenance. Across 1,140 untouched EPL test matches, it beat naive baselines but not normalized bookmaker closing probabilities.

• d9aa36f faststream restore FastAPI 0.140 dependency compatibility

• openai-agents-python retry pre-response WebSocket server errors

• calkit skip unrelated subproject checks for single-item runs

• ag2ai/faststream#2961 adapted FastStream's dependency layer to FastAPI's slotted Dependant model while preserving AsyncAPI metadata and adding regression coverage. It was squash-merged as d9aa36f.

• openai/openai-agents-python#3991 added bounded retry behavior and regression coverage for retryable and non-retryable WebSocket failures.

• calkit/calkit#1028 fixed pipeline selection without weakening relevant validation.

Both changes were reviewed and merged upstream.

• work with me AI engineering, ML engineering, model integration, and reliable automation
• email josh0victor@outlook.com

github.com/dk3yyyy  portfolio →  linkedin →  email →

_{// built with intent. every system a decision.}