Machine Learning Engineer — LLM Fine-Tuning, Alignment & Pre-Training
5+ years training, fine-tuning, and evaluating language models at scale for production systems serving 1M+ users. Based in Bengaluru, India.
- 🧬 Pre-trained a 355M-param GPT-2-medium architecture model from scratch on 28B tokens (DeepSpeed, 4x NVIDIA H100, distributed multi-GPU training)
- 🎯 Fine-tuning (SFT, LoRA, QLoRA) and DPO/preference alignment on Gemma, Llama-2/3, Mistral, Qwen2.5, and GPT models
- 🧪 Improved Phi-4-14B-Instruct by 2% across HF leaderboard benchmarks via Model Stock merging
- ⚡ Architected RAG pipelines (LlamaIndex + Weaviate) with 92% retrieval accuracy and 20% latency reduction
- 🔎 Built a QnA search system (ElasticSearch + BERT + ANN + LTR) serving 1M+ students and 1B+ data points at BYJU'S
- 🏅 Gold medallist (Electrical Engineering) · Top-Rated freelancer, 100% JSS, top 10% on Upwork
| Project | What it does |
|---|---|
| AlgoTutor | AI-powered DSA interview tutor |
| RAG Application | Production RAG pipeline with 92% retrieval accuracy |
| PersonalCourseBuilder | LLM-driven personalised curriculum generation |
| Project Prep AI | AI assistant for explaining your projects in interviews |
| Knowledge Tracing SAINT+ | Transformer model predicting learner mastery over time |
| Quiz Me On | Adaptive quiz generation from any topic |
Python PyTorch HuggingFace DeepSpeed LlamaIndex LangChain CrewAI Weaviate ElasticSearch FastAPI Docker Kubernetes AWS (EC2 · Inferentia-2) GCP
📍 Bengaluru, India · LinkedIn · Open to freelance, contract, and full-time AI roles