Projects Completed
Multi-Document LLM Agent for Q&A using RAG
An advanced Retrieval-Augmented Generation (RAG) solution designed to tackle complex questions that simple semantic similarity-based retrieval cannot solve. This project showcases a sophisticated deterministic ReAct Reasoning as the "brain" of a highly controllable autonomous agent capable of answering wide variety of questions from your own PDF Document.
- ✅ ReAct Resoning: Acts as the "brain" of the agent, enabling orchestration between reasoning and acting.
- ✅ Controllable Autonomous Agent: Capable of answering non-trivial questions from Set of PDF documents
- ✅ Multi-step Reasoning: Breaks down complex Questions into manageable sub-questions.
- ✅ Semantic Chunking: chunks the text based on existing semantics.
- ✅ Flexible Context Retrieval: Uses threshold instead of similarity_top_k which makes the size of the context retrieved in proportional the information needed of inferencing
AI powered - Movie recommender web app
This application has been designed to integrate seamlessly with e-commerce platforms, such as Amazon, or streaming entertainment platforms like Netflix or Spotify, and any website offering items for sale. The fundamental concept behind this project revolves around providing tailored movie recommendations to individual users based on their provided ratings. Imagine you are shopping on Amazon, and you come across items you have previously purchased. You take a moment to rate them before continuing your shopping journey. The application utilizes these ratings to automatically generate movie recommendations aligned with your preferences.
- ✅ Machine Learning based model
- ✅ Deplyed with Flask powered web application
- ✅ Trained with a dataset containing over one million item
Product Reviews Classifier on AWS
This project comprises three key steps aimed at extracting insights from customer product reviews. Firstly, utilizing AWS Glue and Amazon Athena, we've ingested and transformed the dataset, followed by analysis using AWS Data Wrangler to visualize insights. Then i've scrutinized the dataset for biases, generating reports to inform feature engineering and address societal biases. Lastly, employing Amazon Sagemaker Autopilot, i've trained a BERT-based NLP model to classify sentiments in customer feedback, categorizing them into positive, neutral, and negative sentiments.
- ✅ Data Ingestion and visualization
- ✅ Statistical Bias Detection and Data Balancing
- ✅ NLP Classification and Model Deployment
