Mohamed Amrith Project and Contributions

I am an experienced Artificial Intelligence and Natural Language Processing professional, skilled in developing and implementing algorithms and systems.

1. Hello! Iam Amrith Machine learning Engineer

2. Multimodal AI Assistant Project: DAN Project Description: DAN is a multimodal AI assistant designed to answer queries from customer books and text-based PDFs. It retrieves relevant information with references and performs complex reasoning based on the retrieved data. My Contribution: Extracting data from various sources and utilizing visual techniques for data understanding. Developing APIs to process tables within PDFs, enabling DAN to comprehend the data. Integrating with third-party APIs and local language models (LLMs) to enhance DAN's capabilities. Designing and implementing reasoning algorithms to provide comprehensive responses to user queries. Key Technologies Used: Python, Django Framework, API Integration, Multimodal AI.

3. SQL Database Querying and Reasoning System Project Description: Developed a system for answering questions from SQL databases using GPT and sorting algorithms. The user interface allows users to select their company's database and ask questions, generating queries automatically. The system creates a knowledge graph for the database structure and utilizes sorting algorithms to correct character-level mistakes in queries. My Contribution: Developed a sorting algorithm to compare generated queries with the knowledge graph and optimize queries for clarity and accuracy. Utilized Docker to manage GPU resources efficiently for computationally intensive tasks. Implemented reasoning algorithms to make assumptions for sensitive data, leveraging fine-tuned Lama 8 billion model for non-sensitive data and third-party models for sensitive data. Key Technologies Used: GPT, Sorting Algorithms, Docker, SQL, Knowledge Graphs, GPU Computing.

4. Mobile App Project: Image-based Question Answering Project Description: Developed a mobile app capable of answering user questions using images. Users could ask questions directly or upload images for visual queries. The app utilized Stable Diffusion API for image generation and GPT for question answering. For image understanding, LLaMA-3 and Phi-3 models were deployed on AWS and accessed via API My Contribution: Deployed LLaMA-3 and Phi-3 models on AWS for image understanding. Created APIs to facilitate communication with the deployed models. Ensured seamless integration of the models with the mobile app for image-based question answering. Key Technologies Used: AWS, LLaMA-3, Phi-3, API Integration, Mobile App Development, GPT.

5. Dumb Project: Web and Mobile Application Project Description: Dumb is a web and mobile application where users can input various types of media, including images, videos, audio files, PDFs, Excel sheets, PowerPoint presentations, URLs, and more. It is a multimodal project that provides answers and reference media based on user queries or searches. My Contribution: Implementing video-to-text functionality to transcribe audio content within videos. Extracting data from URLs using BeautifulSoup for web scraping. Sending extracted data to be stored in vector databases. Leveraging Google Cloud VM and open-source finetuned models for video and audio transcription, as well as vectorization of data. Key Technologies Used: Python, BeautifulSoup , Google Cloud VM, Multimodal AI, Vector Databases.

6. THANKS FOR YOUR TIME!