Below are some selected projects:
| Repository Name | Description | Year |
|---|---|---|
| llm_reasoning | Research on large language model reasoning for overthinking | 2025 |
| multimodal_cot | Multimodal Chain-of-Thought reasoning in 3 paradigms | 2025 |
| verl | Extension of VeRL to implementation of multi-turn RL for Multimodal Modals | 2025 |
| Repository Name | Description | Year |
|---|---|---|
| VLM-R1 | We Trained VLMs with GRPO to enhance visual question answering | 2025 |
| sft-dpo-rag-training | GalactiTA: Trained a 1B LLM with SFT, DPO, and RAG for scientific question answering | 2024 |
| Coin-segmentation-and-classification | Computer vision project for coin detection and classification | 2024 |
| mountain-car-reinforcement-learning | It focuses on solving mountain car problem with different variations of Deep Q-Network (RL algorithm) | 2024 |
| A-recipe-for-a-successful-tech-review-channel | An in-depth analysis of YouTube Tech channels based on YouNiverse dataset | 2023 |
| datastory | Website of A-recipe-for-a-successful-tech-review-channel |
2023 |
| Repository Name | Description | Year |
|---|---|---|
| llm-efficient-training | Efficient LLM training implementation (Hackathon - 2nd place🎉) | 2025 |
| vlms-for-satellite-hack | AI-powered platform that combines LLM and CV to analyze satellite images (Hackathon - 1st place🎉) | 2024 |
| virtual-me | Virtual "ME", who can answer question about myself | 2024-present |
Deep Learning Frameworks: PyTorch, Transformers (HuggingFace), vLLM, SGLang, VeRL
Data Science: Pandas, NumPy, Matplotlib
HPC Schedulers: SLURM, RunAI