Automating the world with AI Agents. Dedicated to integrating agentic intelligence into every possible system.

Priyanshu Dey  • Peerlist

Adidas AI WhatsApp Assistant is an innovative AI-powered solution designed to enhance customer support for Adidas through WhatsApp. Built using n8n, Twilio, Google Gemini, and Airtable, this assistant automates responses to frequently asked questions, order tracking, product availability checks, and ticket creation. The integration with WhatsApp allows for seamless interaction with customers, providing a 24/7 support system that is both context-aware and efficient.<h3>Key Features</h3><ul><li>WhatsApp Chatbot: Utilizes Twilio for messaging capabilities.</li><li>AI Agent: Powered by Google Gemini for intelligent conversational flows.</li><li>Inventory Management: Real-time lookup and updates via Airtable.</li><li>Order Tracking: Automated tracking and updates through Airtable.</li><li>Customer Support: Efficient ticket creation and issue logging.</li></ul><h3>Workflow Overview</h3><ol><li>Incoming messages are triggered via Twilio.</li><li>Messages are preprocessed for AI analysis.</li><li>The AI agent processes user intent and provides responses.</li><li>Responses are sent back to the user through WhatsApp.</li></ol>This project is open source, allowing developers to contribute and enhance its capabilities further.

FineTuning is an advanced project focused on enhancing medical reasoning capabilities using the DeepSeek-R1-Distill-Llama-8B model. This project employs Low-Rank Adaptation (LoRA) and the Unsloth framework to fine-tune the model on medical reasoning tasks. The implementation is designed to generate structured medical answers with chain-of-thought explanations, making it a valuable tool for clinical applications.<h3>Key Features</h3><ul><li>Advanced Medical Reasoning: Fine-tuned on the FreedomIntelligence medical-o1-reasoning-SFT dataset for enhanced clinical reasoning.</li><li>Efficient Fine-Tuning: Utilizes LoRA adapters for parameter-efficient training, ensuring optimal performance with minimal resource usage.</li><li>Chain-of-Thought Responses: Capable of generating structured medical answers with detailed reasoning traces.</li><li>Quantized Training: Implements 4-bit quantization for improved memory efficiency.</li><li>Experiment Tracking: Integrated with Weights &amp; Biases (WandB) for comprehensive monitoring and analysis.</li><li>Hugging Face Integration: The model is hosted and accessible via the Hugging Face Hub, facilitating easy deployment and testing.</li><li>Inference Scripts: Provides ready-to-use scripts for testing and deployment, streamlining the integration process.</li></ul><h3>Technical Requirements</h3><ul><li>Hardware: Requires an NVIDIA GPU with CUDA support (recommended: 16GB+ VRAM) and 16GB+ system RAM.</li><li>Software: Compatible with Python 3.8 or higher, CUDA 11.8 or higher, and Git for repository management.</li></ul>This project is open source and licensed under the MIT License, encouraging contributions and collaboration from the community.

FineTuning

A2A (Agent-to-Agent) System is a sophisticated multi-agent orchestration platform designed to enable seamless communication and collaboration between specialized AI agents. Built on the Agent-to-Agent (A2A) protocol, this system allows agents to discover, connect, and work together to solve complex tasks. Key Features:<ul><li>Multi-Agent Architecture: Specialized agents for different domains such as Research, Content Creation, Website Building, and Financial Analysis.</li><li>MCP Integration: Model Context Protocol servers for external tool integration, including Weather and Search functionalities.</li><li>Orchestration: A host agent intelligently routes tasks to appropriate specialized agents.</li><li>Multiple Frontends: Offers both a Streamlit web interface and a command-line interface for user interaction.</li><li>Extensible Framework: Easily add new agents and tools to the system.</li><li>Real-time Communication: Utilizes HTTP-based agent communication with session management.</li></ul>The system is open source and licensed under the MIT License, encouraging contributions from the community. It supports dynamic agent discovery and registration, making it a flexible and scalable solution for multi-agent AI systems.

AI Travel Planner is an innovative application that leverages advanced AI technologies to create personalized travel plans. Built with Streamlit, this application offers an interactive user interface that supports trip planning, Q&amp;A, and report downloads. The core functionality is powered by Agentic RAG (Retrieval-Augmented Generation) integrated with a Qdrant MCP vector database for efficient knowledge storage and retrieval. Users can generate comprehensive travel plans that include destination insights, weather forecasts, transport and accommodation recommendations, itineraries, and budgets.Key Features:<ul><li>Agentic RAG with Qdrant for context-aware question answering.</li><li>Multi-Agent Travel Planner for generating detailed trip plans.</li><li>Interactive UI with Streamlit for seamless user experience.</li><li>Hybrid MCP Tools for integrating real-world data.</li><li>MCP Server Control for managing server lifecycles.</li></ul>The application also supports exporting trip plans in Markdown and DOCX formats, and allows users to ask follow-up questions about their trips, enhancing the overall travel planning experience.

WanderSynth

<h3>Overview</h3>AI Doctor with Vision and Voice is an advanced AI-powered virtual medical assistant designed to revolutionize patient interaction. By leveraging cutting-edge technologies, it processes patient voice inputs, analyzes medical images, and generates human-like medical responses. This innovative tool integrates speech recognition, image analysis, and text-to-speech capabilities to deliver a seamless and intelligent medical interaction experience.<h3>Features</h3><ul><li>Speech Recognition: Converts patient voice input into text using OpenAI’s Whisper model via Groq API.</li><li>Image Analysis: Analyzes medical images (like X-rays) using multimodal LLMs to detect abnormalities and suggest remedies.</li><li>AI-Powered Medical Responses: Generates concise, human-like doctor responses based on symptoms and image inputs.</li><li>Text-to-Speech: Converts AI-generated text responses to clear, realistic voice output using gTTS.</li><li>Gradio Web UI: Provides a simple, interactive, browser-based interface for easy user interaction.</li></ul><h3>Technology Stack</h3><ul><li>Python: Primary programming language used for development.</li><li>Gradio: Framework for creating the web interface.</li><li>Groq API: Utilized for multimodal AI models in image analysis and speech-to-text conversion.</li><li>OpenAI Whisper: Employed for accurate speech-to-text conversion.</li><li>Google Text-to-Speech (gTTS): Used for synthesizing text into speech.</li><li>SpeechRecognition: Captures and processes voice input effectively.</li></ul>

AI-Powered-Doctor-with-Vision-and-Voice

Chat with MySQL is an innovative application designed to simplify database interactions by allowing users to query their MySQL databases using natural language. This tool is built on the Streamlit framework, providing an interactive web interface that is both intuitive and user-friendly. By leveraging advanced language models, such as Google Generative AI, the application translates user questions into SQL queries, executes them, and returns results in natural language. Key features include Natural Language SQL Generation, which automatically generates SQL queries from user inputs, and Dynamic Schema Validation to ensure queries align with the database schema. The application also boasts robust Error Handling for SQL syntax and connection errors, and an Interactive Chat Interface for seamless user interaction. The backend is powered by PyMySQL for database connections and SQLAlchemy as a utility wrapper, ensuring efficient and secure database operations. This open-source project, licensed under the MIT License, invites contributions from the community to enhance its capabilities and address any issues.

ChatMySQL

EcomChatBot is an innovative AI-powered product search assistant designed to enhance the e-commerce experience by providing users with detailed product information based on their queries. This chatbot leverages advanced technologies such as LangChain for document processing and vector search, Google Generative AI for natural language understanding, and AstraDB Vector Store for efficient data retrieval. The application is built using Flask, a lightweight web framework, ensuring a seamless and user-friendly interface.<h3>Key Features</h3><ul><li>CSV to Document Conversion: Efficiently converts product data from CSV format into structured documents for better data management.</li><li>Vector Store Ingestion: Utilizes AstraDB Vector Store to store and retrieve product details, enhancing search efficiency.</li><li>AI-Powered Search: Employs Google Generative AI for understanding and processing natural language queries, providing accurate product information.</li><li>Flask Web Interface: Offers a user-friendly chat interface, allowing users to interact with the chatbot effortlessly.</li></ul>With a focus on improving e-commerce interactions, EcomChatBot is an open-source project available under the MIT License, inviting contributions from developers worldwide. Future improvements include implementing authentication for secure API usage, enhancing the UI with modern CSS frameworks, and deploying on cloud platforms like AWS or GCP for broader accessibility.

EcomChatBot

Semantic Book Recommender is an innovative tool that leverages Large Language Models (LLMs) and vector search technology to provide personalized book recommendations. This system processes book descriptions, extracts emotional tones, and categorizes books to enhance discoverability. Key features include:<ul><li>AI-Powered Recommendations: Uses semantic search to find books based on user input.</li><li>Emotional Tone Filtering: Sorts books based on happiness, suspense, sadness, etc.</li><li>Category-Based Selection: Users can filter by genre.</li><li>Interactive Web UI: Built with Gradio for an intuitive experience.</li></ul>Using FAISS for vector storage, Google Generative AI for embeddings, and Gradio for a user-friendly interface, this project enables users to find books that match their interests and emotions seamlessly. The system's architecture includes text data cleaning, semantic search with a vector database, text classification using zero-shot learning, sentiment analysis, and a web application interface.

LLM-Semantic-Book-Recommender

Hate Speech Classification is an advanced project that leverages deep learning techniques to effectively classify text as either Hate Speech or No Hate Speech. This project is built using a robust tech stack that includes popular frameworks and libraries such as Python, TensorFlow, and PyTorch for deep learning, along with Streamlit, Flask, and FastAPI for web development. The project also utilizes Pandas, Numpy, and NLTK for data processing and manipulation.<h3>Key Features</h3><ul><li>Utilizes deep learning architectures for accurate text classification.</li><li>Includes a frontend application built with Streamlit for user interaction.</li><li>Backend logic is implemented using Flask and FastAPI, providing a seamless integration with the frontend.</li><li>Comprehensive project structure with organized directories for configuration, components, and pipelines.</li><li>Licensed under the MIT License, ensuring open-source accessibility.</li></ul><h3>Value Proposition</h3>This project offers a powerful solution for detecting and classifying hate speech, which is crucial for maintaining healthy online communities and platforms. By integrating state-of-the-art machine learning models, it provides reliable and efficient text analysis capabilities, making it an essential tool for developers and researchers in the field of AI and natural language processing.

Hate-Speech-Classification

Kidney Disease Classification is a cutting-edge project designed to predict the health status of kidneys using advanced deep learning techniques. This project leverages the Keras VGG16 architecture to build a robust model capable of distinguishing between healthy kidneys and those affected by tumors. The integration of MLflow and DAGsHub ensures efficient experiment tracking and version control, making the system both scalable and reproducible.<h3>Key Features</h3><ul><li>Deep Learning Model: Utilizes the VGG16 architecture for high accuracy in classification tasks.</li><li>Experiment Tracking: MLflow is employed for logging experiments and tracking model performance.</li><li>Version Control: DAGsHub provides a platform for experiment versioning and collaboration.</li><li>Data Management: DVC is used for lightweight experiment tracking and pipeline orchestration.</li><li>Deployment: The model can be deployed as an API or web app using Streamlit.</li></ul><h3>How to Run</h3><ol><li>Clone the Repository: Use the provided GitHub link to clone the project.</li><li>Create and Activate Conda Environment: Set up a Python environment using Conda.</li><li>Install Dependencies: Use the requirements.txt file to install necessary packages.</li><li>Run the Application: Deploy the backend with Python and the frontend with Streamlit.</li></ol>This project represents a significant advancement in the field of medical AI, offering a reliable tool for kidney disease classification.

Kidney_Disease_Classification

BreastCancerAI is a machine learning-based application designed to assist medical professionals in diagnosing breast cancer. Utilizing the Wisconsin Breast Cancer Dataset, this application predicts whether a breast mass is benign or malignant based on a set of cellular features. The project includes a Python-based model training script and a Streamlit web application.<h3>Key Features</h3><ul><li>Data Cleaning and Preprocessing: Handles missing values, removes unnecessary columns, and maps diagnosis labels to binary values.</li><li>Feature Scaling: Utilizes StandardScaler to normalize feature values.</li><li>Model Training: Trains a Logistic Regression model to classify breast masses with high accuracy.</li><li>Interactive Web Application: Allows users to input cellular measurements using sliders, displays predictions with probability scores, and visualizes input features using radar charts.</li></ul><h3>Dataset</h3>The application uses the Breast Cancer Wisconsin (Diagnostic) Dataset, which includes 569 instances and 32 features. The target variable is the diagnosis, with benign labeled as 0 and malignant as 1.<h3>Requirements</h3><ul><li>Python 3.8</li><li>Dependencies: numpy, pandas, plotly, scikit_learn, streamlit, altair</li></ul><h3>Usage</h3>Clone the repository, download the dataset from Kaggle, and place it in the Data/ directory as data.csv. Train the model by running the script and launch the Streamlit application to interact with the web interface.<h3>License</h3>This project is licensed under the MIT License.

BreastCancerAI

Delivery Time Predictor is an innovative tool designed to enhance the efficiency and transparency of food delivery services. By leveraging machine learning models, this project aims to accurately estimate delivery times based on historical data and various factors such as distance, traffic conditions, and weather. This open-source project is hosted on GitHub and is available under the MIT License.<h3>Key Features</h3><ul><li>Real-time Prediction: Users can input order details to receive an estimated delivery time instantly.</li><li>Batch Prediction: Allows uploading a CSV file for bulk predictions, making it ideal for businesses handling multiple orders.</li><li>Model Training: Users can train a new model using historical data to improve prediction accuracy.</li><li>Logging and Exception Handling: Provides detailed logs for debugging and monitoring, ensuring smooth operation.</li></ul><h3>Technologies Used</h3><ul><li>Python: The core programming language used for development.</li><li>Flask: A lightweight web framework for building the application's web interface.</li><li>scikit-learn: Utilized for implementing machine learning models.</li><li>pandas &amp; NumPy: Essential for data handling and manipulation.</li><li>Pickle: Used for model serialization, allowing easy storage and retrieval of trained models.</li></ul>This project is a valuable resource for developers and businesses looking to integrate AI-driven delivery time predictions into their operations.