Henrik Sachdeva

Hey! I’m Henrik, a Computer Science undergraduate at Simon Fraser University specializing in LLM fine-tuning, alignment, and retrieval-augmented generation.

I build and evaluate end-to-end NLP pipelines using PyTorch, Hugging Face, and FAISS, achieving measurable gains on benchmarks such as HotpotQA and E2E Table-to-Text.

Technical Skills

• DL/ML Frameworks: PyTorch, Hugging Face (Transformers, TRL, Datasets), Scikit-learn, numpy, and pandas.
• LLM Techniques: RAG (Retrieval-Augmented Generation), ORPO (Preference Optimization), and PEFT (LoRA, Prefix Tuning).
• Models: Experience with T5/FLAN-T5, BERT, GPT-2, and Qwen 2.5.

Featured Projects

• RAG for Multi-Hop QA: Architected a two-stage Retriever pipeline using BGE Sentence Embedder and FAISS, achieving a Joint F1 of 33.91% on HotpotQA.
• ORPO Preference Optimization: Applied the Odds Ratio Preference Optimization algorithm to Qwen 2.5, reducing training parameters by 90%+ using LoRA.
• Prompt Tuning: Utilized Prefix Tuning on distilgpt2 to achieve a BLEU score of 30 on the E2E Table-to-Text task.
• Journeyhacks Coding Platform: Created a real-time competitive coding game with live API integration and leaderboard.

Education

Bachelor of Computer Science (AI & ML Concentration)
Simon Fraser University | Graduating April 2026
Relevant Coursework: Deep Learning, NLP, Machine Learning, and Data Structures & Algorithms.