CS undergrad at VTU, Bengaluru expected 2027. I write the parts most people import — rate limiters, RAG pipelines, SQL safety layers, a storage engine — because I want to know exactly how the thing under the thing works.
Most of my projects start from the same irritation: a framework hides the interesting part. So DocuMind isn't a LangChain demo, it's a RAG pipeline where I chose the chunk sizes, tuned the retrieval, and cut response latency with Redis myself. SchemaSense doesn't just call an LLM to write SQL — it validates and blocks destructive queries before anything touches a real database.
That instinct is why I'm currently writing a storage engine from scratch in Go — slotted pages, a buffer pool, B+ trees, a write-ahead log with crash recovery — the layer underneath the databases I've already built products on top of.
Outside of backend work: space, rockets and propulsion systems, literature, badminton, and slowly learning Spanish.
Slotted pages, buffer pool with LRU eviction, B+ tree with full split/merge rebalancing, and a write-ahead log with CRC32 checksums for crash recovery. Benchmarking against BoltDB.