"learning begins with the death of ego" - liber primus
first-year computer science undergraduate focusing on low-level distributed infrastructure, concurrent systems execution, and ML inference optimization frameworks.
currently working on : a method to optimize artificial brain design with "smart" neurons
languages: c, go, python, c++, typescript, javascript, assembly
frameworks: grpc, tensorflow.keras or pytorch, next.js (app router) + tailwind css, socket.io or pubnub, SDL3/pygame (i guess), discord.py, flask, selenium
infra: supabase, neon, redis, mongodb, docker, firebase, vercel
- vllm-project/vllm: merged #54265 (docs example for Renderer.render_cmpl()); opened RFC #57672 proposing RSQR, a KV-cache eviction design eliminating compounding rotation drift and decreasing latency
- binary/web exploitation & reverse engineering (stack/rop layers)
- bit of crypto
- algorithmic problem solving (cses tracker)
- How I Built RoommateFinder (and Optimized It With Go) - building a full-stack matching platform from scratch; Go concurrency, caching, fan-in/fan-out worker pools, and benchmarking the wins
- A Custom x86 Mini Assembly Emulator (and Why I Made It) - writing an x86 emulator in C from first principles: addressing modes, the stack, FLAGS register, JMP/CMP, Turing completeness
- dLLM: An Async Pipeline-Parallel LLM Orchestration Framework - exploring speculative decoding and block-wise quantization for LLM inference; documents where the design failed and why
- RSQR: A New Perspective on Efficient KV-Cache Eviction for Streaming LLMs - designing a RoPE re-rotation scheme for KV-cache eviction, diagnosing a numerical drift bug, and proposing it as an RFC to vLLM

