Portfolio
WA Creel
Transforms WDFW creel reports and NOAA current data into actionable fishing insights
View Case Study ↓
Problem
WDFW creel reports and NOAA current predictions exist as raw tabular data. Anglers lacked tools to transform this data into actionable decisions about where and when to fish.
Solution
Built analytics platform that processes WDFW creel survey data and NOAA current predictions into actionable fishing insights. Key features:
- Multi-metric analysis: CPUE (efficiency), Success Rate (reliability via Poisson), Hotness Score (composite ranking)
- Statistical significance testing: Expands date windows until sufficient sample sizes ensure reliable predictions
- Trip Planner wizard: Personalized recommendations based on species/area/timespan preferences with multi-criteria scoring
- NOAA current integration: Slack current predictions using multi-station averaging for optimal fishing windows
- Regulation filtering: Automatically excludes closed seasons from recommendations
Tech Stack
- Backend: Python with custom HTTP server, SQLite + GCS persistence
- Frontend: Vanilla JavaScript with Chart.js for time-series visualizations
- Infrastructure: Docker on GCP Cloud Run with automated CI/CD via Cloud Build
- External APIs: WDFW creel data, NOAA current predictions, WDFW regulations
Outcome
Production fishing analytics platform serving Puget Sound recreational anglers. Statistical methods (Poisson-based success rates, date window expansion for significance) ensure recommendations meet minimum confidence thresholds. Trip Planner generates weighted recommendations filtered by current regulations, improving catch success through data-informed decision making.
claude-code-search
Semantic search layer for Claude Code to navigate unknown projects with minimal token usage
View Case Study ↓
Problem
Claude Code wastes tokens reading entire codebases to find where functionality lives. In unknown projects, “Where does X happen?” triggers broad file scans that consume context budget without guarantees.
Solution
Semantic search layer using ChromaDB vector embeddings. Code chunked by class and function definitions—right granularity for single queries. File watcher monitors project directory, auto-updates index on any change (Claude, user, or external). Claude Code queries in natural language → vector search returns ranked chunks → read only relevant files. Token savings over speed.
Tech Stack
- Vector Store: ChromaDB for persistent semantic index
- Embeddings: Code-aware models via HuggingFace Transformers
- Language: Python
- Integration: CLI/API for Claude Code workflows
Outcome
Narrow file set before reading. Claude Code finds functionality in unfamiliar repos without burning tokens on full codebase scans. Main win: token efficiency, not time.
api-gateway-hub
Backend-for-Frontend API aggregator with Redis caching and resilient retry logic
View Case Study ↓
Problem
Frontend applications making multiple external API calls create latency bottlenecks and expose API keys. Direct client calls lack caching, retry logic, and rate limit protection.
Solution
Built BFF (Backend-for-Frontend) integration service aggregating 3 external APIs (weather, cryptocurrency, country data) with intelligent Redis caching, exponential backoff retry, and stale-cache fallback on failures.
Tech Stack
- Framework: FastAPI 0.115 with async/await for concurrent API calls
- Caching: Redis 7.2 with tiered TTLs (5-24h based on data freshness)
- Resilience: httpx + tenacity for automatic retry with exponential backoff
- Database: PostgreSQL 16 for request logging and analytics
Outcome
Single unified endpoint reduces frontend complexity and latency. Redis cache-aside pattern with stale fallback achieves 95%+ cache hit rate. 84% test coverage with testcontainers for real Redis/PostgreSQL integration tests.
authkit
Production-ready authentication microservice with JWT and OAuth2 social login
View Case Study ↓
Problem
Building secure authentication from scratch risks common vulnerabilities (weak password hashing, token leaks, OAuth misconfiguration). Third-party services like Auth0 add cost and vendor lock-in.
Solution
Built self-hosted authentication microservice with dual-token pattern (15min JWT access + 7-day refresh tokens), OAuth2 social login (Google/GitHub via Passport.js), and rate-limited endpoints. Drop-in replacement for Auth0/Clerk.
Tech Stack
- Runtime: Node.js 20 LTS + Express 5 with TypeScript 5.6 strict mode
- Authentication: jsonwebtoken for stateless JWTs, Passport.js for OAuth2
- Security: bcrypt password hashing, Zod validation, express-rate-limit
- Database: PostgreSQL 16 with referential integrity and cascade deletes
Outcome
Comprehensive test suite (100+ tests, >70% coverage) with testcontainers for real database integration tests. Docker Compose deployment ready. Eliminates auth vendor costs while maintaining security best practices.