Devananda Sreekanth@devu

ChatmanGPT revolutionizes development by automating code generation, aiding in rapid prototyping, and optimizing code based on AutoGPT. It's a comprehensive system builder with unique features like Prompt Matrix Innovation and real-time evaluation.

Raga Music Generation Pipeline: RagaCraft Our project, RagaCraft, bridges the gap between raw human emotion and the timeless art of raga music using cutting-edge AI. Here's a deeper dive into the underlying process: Customer Interaction: Users interact with our platform, sharing their current emotions and contextual information. For example, "I am feeling romantic today. It is Valentine's Day. I'd like a song to suit the mood." JavaScript Selection: Our system, powered by JavaScript, scans the user's input to select an appropriate raga that resonates with the given emotion. OpenAI Integration: To add depth and specificity, RagaCraft sends a refined request to OpenAI: "Generate a text-to-music prompt for a single romantic raga. Include parameters such as tempo, scale, pitch, and rhythm to optimize the romantic mood. Define ideal values for these features." OpenAI's Response: The API, enriched with musical knowledge, replies with precise musical direction. For instance, "For a romantic setting, employ the Hindustani raga Kamboji. Utilize a medium-slow tempo, major scale, and a high pitch with low undertones. The rhythm should be gentle with a 4/4 signature. Dynamics can vary, with crescendos and decrescendos, ensuring a light texture and smooth timbre." Audiogen Transformation: The detailed prompt from OpenAI is fed into Audiogen, which processes it and crafts a song that encapsulates the user's emotions. Delivering the Experience: Our user interface then presents the generated raga song to the user, completing a journey from raw emotion to personalized musical expression. Through RagaCraft, we're redefining the way users experience and interact with traditional music forms in the age of AI.

AISmartTask: Coded using Open-interpreter, AISmartTask is an innovative AI-driven task management solution tailored for coders or anyone who likes to manage daily routines. Leveraging the power of OpenAI's text engines, this tool is integrated with a Telegram bot, offering an intuitive user experience. Key features include dynamic task management that allows for task addition, prioritization, and AI-powered suggestions. Its document parser extracts essential summaries and highlights critical keywords. Additionally, AISmartTask visualizes task priorities, enabling users to get a clearer view of their tasks at hand. One standout quality is the tool's learning modules, which web scrape for knowledge, encouraging continuous learning. Users can interact with the system via intuitive Telegram commands (vocal or text) and provide feedback for continuous system improvement. With a modular Python architecture, AISmartTask is not just a task manager; it's a revolutionary tool aiming to optimize and enhance the daily routines of coders. It is scalable to any user and other applications such as enterprise project management, academic tasks management (students and educators), task management in healthcare, in the freelance and gig economy, or to manage tasks in any use cases.

With 15% of working-age adults facing mental disorders and an annual loss of US$ 1 trillion in the world due to impaired productivity from depression and anxiety, the necessity for real-time emotional and physiological monitoring is paramount. PsychGen, developed using AutoGen multi-agent platform, aims to democratize mental healthcare through AI-driven personalized psychotherapy sessions, addressing the global rise in mental health issues and the inaccessibility of traditional therapy. Utilizing a robust technology stack including AutoGen, Streamlit, OpenAI, AssemblyAI, Audiocraft, and DALL-E, PsychGen transitions seamlessly from text to audio, audio to music, and finally to video to enrich the therapeutic experience. The project encompasses multi-faceted agents such as a Psychological Chatbot for profiling and script generation, Speech and Music Generation for a meditative voice output and soothing soundtrack, Image and Video Generation for visual representation of therapy themes, and a Mixing agent for creating a comprehensive therapy session. PsychGen holds the promise of democratized access to psychotherapy, real-time personalized mental health support, and scalability for integration with other mental health resources, envisioning a future where quality mental healthcare transcends traditional barriers and is accessible to all.

ResearchWriterGPT is an Advanced Multimodal Research Paper Writing Assistant is a groundbreaking tool designed to transform academic writing. It harnesses the language and vision capabilities of GPT-4 to assist in crafting research papers, processing both textual and visual data to ensure thorough coverage from abstract to conclusion in APA format. The project showcases its multimodal capabilities, including image and chart scanning and analysis. It offers direct access to academic databases like Google Scholar, Semantic Scholar, Pubmed, etc., facilitating the literature review process by aggregating and filtering peer-reviewed information. Additionally, the tool enhances user experience through interactive dialogues, audio interaction, PDF analysis, and PPT downloading option. GPT-4 Vision expands the application's scope by enabling detailed image analysis, such as reading texts and interpreting charts from research photos and medical images. The integration of a Pinecone-based RAG system allows users to upload a collection of documents, which the system appends to for relevant response generation. This creates a vast knowledge base, potentially processing millions of articles for quick, contextually relevant responses, supporting efficient document management and advancing context-based AI LLM feed. TruLens further strengthens the tool by evaluating hallucinations in three key dimensions: context relevance, groundedness, and answer relevance. This ensures the LLM application is free from hallucinations, delivering accurate and relevant information. The Trulens leaderboard feature displays the relevancy score of the LLM responses giving realtime feedback. Future expansions aim to incorporate advanced models for face sentiment analysis and object detection, predictive bibliography features, and comprehensive writing support covering all aspects of a research paper.

This project explores the development of an AI-based quiz generation system using GPT-3.5 and Vectara to foster personalized and adaptive learning experience. Our methodology involved sourcing relevant structured dataset, data pre-processing, embeddings generation, vector database storage, hybrid-search retrieval, LLM feed, prompt engineering, and context-based response. The primary challenge addressed is the insufficient customization in quiz generation and overcoming the challenge of generating precise and contextually relevant quiz questions that minimize the risk of LLM generating incorrect or misleading information (hallucinations). We also have included an option for the user to upload their preferred document and let the app generate quiz. As the users answer the questions, the LLM will generate further questions adapting the difficulty level. The hallucination levels are found to be really low as per Vectara in-built evaluations. This is designed for interactive AI based user learning and is scalable to different arenas like education, online learning, and employee trainings.