starone01.me · linkedin · x @iamstarone01 · ping@starone01.me
final-year EEE @ Coimbatore · learning LLVM / MLIR
// pipeline.mlir — how I think about building
func.func @starone01(%idea: !prod.prototype) -> !prod.shipped {
%ast = "frontend.parse"(%idea) : (!prod.prototype) -> !ir.ast
%lowered = "midend.lower"(%ast) : (!ir.ast) -> !mlir.module
%opt = "midend.canonicalize"(%lowered): (!mlir.module) -> !mlir.module
%bin = "backend.emit"(%opt) {target = "prod", opt = "-O3"} : (!mlir.module) -> !prod.shipped
return %bin : !prod.shipped
} // scroll ↓ to execute each passI'm Prashanth, StarOne01. Final-year EEE student interested in where hardware meets software: compilers, inference, and systems around AI.
Started coding on a phone with Termux. Since then I've helped build a clinical AI system, made some small LLVM contributions, and tried out embeddings and multilingual LLM work. Currently learning MLIR.
| project | what it is | |
|---|---|---|
medclara |
Medclara — Clinical AI | Voice-first multilingual clinical docs. IndicConformer + Gemma for SOAP. Worked on arch + infra. |
movieslikethis |
MoviesLikeThis — Embeddings | Match films by feeling, not genre. Small custom embeddings experiment. |
sherlock_sft |
Sherlock SFT — Fine-Tuning | Character LM via SFT. Spent a while tracking down a QLoRA NaN loss. |
bfloat16 |
bfloat16 — Numerics | Small C++ bfloat16 implementation. |
phrasenux |
PhraseNuX — origin | C++ CLI password manager, AES, zero deps. Early project, written on a phone. |
Languages → Go · Python · C++ · TS / Next.js
Systems → LLVM · MLIR · Compilers · Local Inference · CUDA (learning)
ML / AI → SFT / QLoRA · ASR · Embeddings · RAG
Infra → AWS · GCP · Postgres · Qdrant · Fedora
- Learning MLIR dialects + passes, building on X86 + Sema work
- Playing with local LLM inference, Ollama internals