AI/ML engineer-in-training building inference systems from scratch. B.Tech CS (AI & ML), Jain University, Bengaluru — graduating 2027.

I care about understanding systems down to the metal rather than gluing together APIs. That shows up as a preference for local-first, privacy-first, zero-external-dependency engineering — from data prep through inference serving, in industrial environments where data never leaves the perimeter.

A from-scratch LLM inference server, and my main portfolio project:

• QLoRA fine-tuning pipeline on Qwen3-27B for domain adaptation
• Speculative decoding with rejection sampling — draft model distilled via sequence-level KD against the fine-tuned target's own generations
• Paged KV-cache implemented via Triton
• Continuous batching — benchmarked throughput scaling ~6x from one to eight concurrent users

• Predictive Ghost-Text Daemon (Ideation) — always-on autocomplete via a resident Unix-socket inference process, sub-100ms latency budget, prefix caching, and shell-history personalization through a fine-tuning pipeline with secrets-scrubbing
• prr — a self-hosted code-review bot pairing an Ollama model with ruff, mypy, and bandit; results normalized through a typed Pydantic schema, posts inline PR comments on GitHub

Python · Java · SQL · PyTorch · TensorFlow · scikit-learn · XGBoost · CUDA · Triton · MLX · LoRA/QLoRA · speculative decoding · paged KV-cache · continuous batching · RAG (FAISS/ChromaDB) · FastAPI · Docker · Ollama · Jenkins · Git/GitHub Actions

Theory foundation: constrained optimization, KKT conditions, multi-armed bandits.

AI/ML engineering internships with a path to full-time — especially roles where inference performance, model serving, or applied ML infrastructure are the actual job, not an afterthought.

• LinkedIn: likhith-v