Experience

• 

  Xueyong educational consultancy, dengdi educational technology

  Apr 2016 - Jun 2023

  Entrepreneur, developer, teacher/tutor

  Developed 10 educational technology apps, including websites, mobile apps, VR classrooms, interactive idea-generating apps, an online exam system, educational graphs, and animation videos.Taught UK A-level Economics, Maths, Further Maths, Physics, Accounting, IELTS Speaking and Writing, and undergraduate-level Introduction to Python and R.Tutored data science and programming at the undergraduate and graduate levels, including courses at University College London (UCL), Columbia University, Imperial College London, and Toronto University.Graduate-level economics courses, including portfolio theory, the capital asset pricing model and efficient markets, options I and II, forwards, futures and corporate risk management, capital structure I and II.Undergraduate courses, including quantitative analysis in management, financial statistics, and advanced financial reporting. Show less

• 

  Skinopathy

  Sept 2024 - Dec 2024

  Student ai researcher - fine-tune openai whisper large-v3 turbo

  * Fine-tuned the OpenAI Whisper Large-v3 Turbo model to improve audio recognition of dermatology terminology using QLoRA.* Achieved a 33% reduction in Word Error Rate (WER), improving accuracy from 14.5% to 9.5%.* Cleaned and processed over 100 dermatology documents, which were used to synthesize more than 780 English audio samples with varied accents and genders using the Google Text-to-Speech API for training purposes.

• 

  Kaggle

  Sept 2024 - Dec 2024

  Kagglex fellowship - fine-tune gemma-2b, gemma-7b; fintech

  - Selected as one of 81 fellowship recipients from over 3,000 applicants for the KaggleX fellowship program.- Final project ReguGuard AI selected as one of the 16 final showcases in the end of program celebration- Built an AI chatbot ReguGuard answering financial risk compliance questions by fine-tuning Gemma-2b and Gemma-7b LLMs.- The best finetuned model, gemma_7b_finRisk_r6_4VersionQ, achieved accuracy rate of 78.6%.- Tested effects of model sizes (2b vs 7b), number of versions of similar questions (3 vs 4), LoRA rank sizes (6 vs 10), and beam search of 3.- Adapted LLamaIndex's RAFT DatasetPack module and used OpenAI API to generate over 14,000 question-answer pairs for training- Show less