what i'm building
an open source graphql caching solution with customizable middleware and redis integration. boosts api performance while keeping devs in control.
what i did
engineered a custom TTL + LRU caching layer in Express.js and TypeScript, improving query throughput and reducing resolution latency by 63% (120ms → 45ms) under load
integrated Redis with tag-based invalidation to maintain cache consistency and ensure correctness across distributed instances
refactored middleware for extensibility and developer experience, enabling devs to customize caching strategies with minimal overhead
benchmarked system under simulated load, demonstrating 90%+ cache hit rates and resilient failover behavior
stack
node.js · express · typescript · redis · graphql
an evaluation pipeline for multi-candidate LLM outputs with observability-first design. benchmarks completions at scale, reranks by reasoning quality, and logs results for reproducibility.
what i did
built a fault-tolerant evaluation pipeline in Python with CLI entrypoint for batch runs
integrated PyTorch + Hugging Face embeddings to generate reasoning vectors and score completions
added structured JSON logging and OpenTelemetry tracing spans for step-level observability
scaled to 10K+ prompts using Hugging Face datasets, with reproducible JSON-logged runs
stack
python · pytorch · hugging face · json · opentelemetry · fastapi