Pinned Loading

1. kiln kiln Public

   A full-stack LLM inference engine built from scratch in C++/CUDA with a Python control plane — paged KV caching, continuous batching, INT8/INT4 quantization, speculative decoding, tensor parallelis…

   C++

2. aegis-observability aegis-observability Public

   Finds which service actually caused an outage, not just the loudest one. Correlates Kafka telemetry on event-time watermarks, builds a dependency graph from span parentage, and isolates root cause …

   Python

3. cleartext-api cleartext-api Public

   Async inference platform for a slow BERT classifier: FastAPI enqueues to Celery, an in-process micro-batcher merges concurrent requests into one forward pass, and a model-versioned Redis cache, cir…

   Python

4. kv-store kv-store Public

   LSM-tree key-value store in Python — write-ahead log, memtable, SSTables with Bloom filters, size-tiered compaction. Crash-safe and verified with a real kill -9 recovery test. Redis-style TCP proto…

   Python

5. go-auth-service go-auth-service Public

   JWT identity service in Go. Rotating refresh tokens with reuse detection — replaying a consumed token revokes every session — plus Redis-backed revocation and rate limiting that degrades to in-memo…

   Go

6. springboot-resource-api springboot-resource-api Public

   Spring Boot resource API that owns its data but not identity — every request is validated against a Go auth service over gRPC. Token-hash-keyed Caffeine cache (60s TTL) and a Resilience4j circuit b…

   Java