YouTube Live Stream Bot

Description: A Chrome Extension designed to automate interactions within YouTube Live chat, leveraging AI for intelligent response generation and user engagement.

Key Responsibilities & Achievements:

• Developed the Chrome Extension foundation using JavaScript, Chrome APIs (e.g., `chrome.runtime`, `chrome.tabs`), and asynchronous programming patterns for non-blocking operations.
• Integrated YouTube Data API for real-time fetching of live chat messages and programmatically posting responses.
• Incorporated OpenAI API (e.g., GPT models) for generating contextually relevant and engaging AI-driven chat replies.
• Engineered AI features including conversational history management using `chrome.storage` for persistence and applied prompt engineering techniques to improve context understanding and response quality.
• Implemented secure Google OAuth 2.0 flow using `chrome.identity` API for user authentication and authorization to access Google services.
• Designed and implemented robust error handling and logging mechanisms for external API calls and extension operations.

Tech Stack: JavaScript, Chrome Extension APIs (Storage, Identity, Runtime, Tabs), YouTube Data API, OpenAI API, Google OAuth, HTML, CSS, Asynchronous JavaScript.

Chatbot Development (Atlas Copco)

Goal: Create a chatbot to answer product-related questions based on provided data, managing token limits for GPT models.

Solution: Utilized GPT-4 with data from Excel (product info: Model Name, Price, Capacity, Links). Developed a Flask-based interface. Implemented strategies to handle GPT token size constraints effectively.

Tech Stack: OpenAI GPT-4, Flask, openpyxl, Pandas, NumPy, NLTK, Scikit-learn, TensorFlow, Keras.

View Project

Fraud Transaction Detection

Goal: Build an ML model to detect fraudulent transactions from historical data.

Solution: Performed EDA and Feature Extraction. Applied and compared multiple ML algorithms (XGBoost, SVC, Logistic Regression). Achieved 90% accuracy with XGBoost after hyperparameter optimization. Deployed the model for production.

Tech Stack: Scikit-learn, XGBoost, Flask, Pandas.

Michel AI Parabroker ChatBot (Cynario.ai)

Goal: Scrape data from various lenders and enable users to query this data effectively.

Solution: Implemented Retrieval Augmented Generation (RAG) for searching. Used Pinecone for vector data storage and Cohere Reranking to refine search results from scraped PDF and web data.

Tech Stack: PyPDF2, BeautifulSoup, Gunicorn, GPT-4, Pinecone, RAG, Cohere, Flask.

View Project

Football Analyst ChatBot (Kaizenwin)

Goal: Create a chatbot to analyze football pass data for specific teams/players and predict future match outcomes.

Solution: Scraped data from diverse football sources. Leveraged RAG, Pinecone, and Cohere to process and retrieve insights from the large dataset.

Tech Stack: BeautifulSoup, PyPDF2, RAG, Pinecone, Cohere.

View Project

Image Generation (Aieventbooth)

Goal: Allow users to generate images based on presets applied to a given input image.

Solution: Used Stable Diffusion for image generation. Trained the model on custom images. Image generation is triggered by user-defined presets.

Tech Stack: Transformers, PyTorch, Gunicorn, Stable Diffusion.

View Project

AI Energy Tender Matching

Goal: Match incoming tenders against an organization's privacy policies and internal data, explaining the rationale for matches.

Solution: Chunked organizational data and fine-tuned models (Llama-2-7B/13B, Mistral-7B). Used cosine similarity for initial matching and GPT-4 for detailed explanations of matched data.

Tech Stack: Transformers, PyTorch, Gunicorn, Llama-2, Mistral, GPT-4.

View Project