Just a student messing around with AI and building ML models.
Half learning, half experimenting, and occasionally making something that actually works. 🚀

Tech Stack: Python, PyTorch, Hugging Face Transformers, ERRANT, Pandas, LaTeX

• Conducted a comparative analysis of Transformer architectures (T5, Flan-T5, BART) for sequence-to-sequence error correction.
• Implemented a two-stage fine-tuning curriculum:
  • Domain Adaptation: Trained on 1M+ noisy sentences (Lang-8).
  • Specialization: Refined on high-quality gold corpora (NUCLE, FCE, W&I).
• Implemented an existing Edit-based System Combination (ESC) to ensemble model outputs, leveraging the complementary strengths of precision-focused (T5) and recall-oriented (BART) models.
• Achieved a top F0.5 score of 68.97 on the BEA-2019 benchmark with T5-base, outperforming the significantly larger BART-large model.

Tech Stack: Jupyter Notebook, Apify API, Gemini API

• Built a multi-classification model to classify Google reviews into: ad, spam, irrelevant, rant_no_visit, relevant.
• Fine-tuned DistilBERT to improve review filtering efficiency.
• Processed and cleaned data, scraped reviews with Apify API, and trained the model with high-quality inputs.
• Evaluated performance with accuracy and F1 scores, ensuring reliable classification.

Tech Stack: React Native, DeepSeek API, TypeScript, Python

• Developed during a hackathon to provide personalised travel planning.
• Features:
  • Dynamic itinerary generation based on user inputs (location, budget, preferences, dates).
  • Chatbot for real-time itinerary modifications.
  • Speech translation to enhance accessibility.
• Adapted to real-world conditions (e.g., crowds, weather) to improve user experience.

Tech Stack: Excel VBA

• Semi-automated rostering system.
• Features:
  • Roster generation algorithm with staff swapping support.
  • Analysis report generation and sheet protection.
• Reduced rostering time from two weeks → under one minute, boosting efficiency and accuracy.
• Conducted user interviews and integration testing to ensure seamless adoption.

Tech Stack: React Native, Express.js, Node.js, Axios, MongoDB, JavaScript, eBay API

• Full-stack marketplace app.
• Key features:
  • Buy, list, and sell items with user authentication and payments.
  • eBay API integration for price comparison.
  • Basic image recognition with suggested item names.