Engineering Solutions with Measurable Impact
Self-operating cryptocurrency market data ingestion from 5 exchanges (Binance, Coinbase, Kraken, OKX, Bybit). Built on TimescaleDB with intelligent backfill, auto-recovery, and checkpoint tracking.
Full-stack app using stylometric fingerprinting (burstiness, perplexity, rare-word usage) to detect AI vs. human writing. Features heat map visualization and DSPy-powered style rewriting.
Published npm package (@dustinober/glm-monitor) for real-time tracking of LLM API tokens, model calls, and quota limits. Features auto-refresh, historical charts, and CSV export.
A hybrid heuristic + Google OR-Tools solver that generates optimal multi-constraint schedules in seconds. Reduced administrative scheduling time by ~80% in pilot tests.
Multi-agent system (LangGraph) that maps global news to supply chain graphs via Neo4j using NER. Real-time risk detection allows for 40% faster contingency planning.
Uses DSPy to programmatically optimize prompts based on user feedback. Increased AI grading alignment with human instructors from 65% to 92% over 50 iterations.
A CLI assistant with 40+ specialized commands for refactoring and testing. Benchmarked a 30% reduction in typical "chat-to-code" cognitive context switching overhead.
Streamlit dashboard using BERTopic for grouping and VADER for sentiment scoring. Provides actionable daily intelligence summaries in < 5 minutes of review.
Pipeline using vision transformers to convert document screenshots directly into clean Markdown. High-fidelity extraction of academic papers for ingestion into RAG pipelines.
These projects represent my hands-on engineering work across AI/ML, optimization, and full-stack development. Each project follows a Problem → Solution → Impact framework, demonstrating measurable outcomes and production-ready architecture patterns.