student

Abhinav Sunil  • Peerlist

PicSage AI is a web-based application that allows users to upload any image — from anime and manhwa to cartoons and celebrities — and receive context-aware recognition along with real-time answers to questions about the image. Leveraging AI models for image understanding and integrating DuckDuckGo for up-to-date web data, the system provides accurate, interactive insights.

PicSage AI – AI-Powered Image Recognition & Q

A lightweight, no-config, plug-and-play Python library for: ✅ Text Generation ✅ Summarization ✅ Sentiment Classification ✅ Translation (200+ languages!) ✅ CPU-friendly & fully offline — no API keys, no cloud, no drama.

genai-textkit

Developed a web-based application to predict customer churn using an Artificial Neural Network (ANN) built with TensorFlow. The app, deployed on Render using Streamlit, enables real-time predictions through an interactive user interface, allowing businesses to identify at-risk customers and optimize retention strategies. Key features include data preprocessing, model training, and dynamic visualizations. Technologies used: Python, TensorFlow, Keras, Streamlit, Pandas, NumPy, Scikit-learn, and Render for cloud deployment. Achieved [e.g., 85% model accuracy, if applicable]. Explore the live app: https://customer-churn-modelling-using-ann.onrender.com

Customer Churn Prediction with Artificial Neu

github link: https://github.com/IamAbhinav01/Next-Word-Prediction-Using-LSTM Built and deployed an LSTM-based model to predict the next word in a sentence using conversational text data. The model was trained on tokenized sequences from the Cornell Movie Dialogues Corpus, learning contextual relationships in natural language. A Streamlit-based web interface allows real-time predictions from users. Key Highlights: Used sequence padding and one-hot encoding for training Trained for 5 epochs using categorical cross-entropy loss Achieved 18.46% validation accuracy with val_loss: 5.1

Next Word Prediction Using LSTM

Developed a sentiment analysis model using a SimpleRNN architecture to classify movie reviews as positive or negative. Implemented text preprocessing techniques such as tokenization and padding. Deployed the model with a user-friendly Streamlit web app interface allowing users to input custom reviews and receive instant sentiment predictions. Improved model accuracy by tuning hyperparameters and experimenting with embedding layers.

IMDb Movie Review Sentiment Classifier

Fine-tuned Google's gemma-2b large language model using LoRA (Low-Rank Adaptation) for legal domain instruction tuning. Trained on structured Indian legal instructions and responses, optimized for low-resource environments using 8-bit quantization (bitsandbytes). Implemented evaluation, deployment, and inference using Hugging Face's transformers, PEFT, and Trainer APIs. Uploaded and documented the final model on the Hugging Face Hub with a reproducible config and tokenizer.

Fine-Tuning Gemma-2B with LoRA for Legal Text

ust wrapped up building SketchStorm, a Claude-compatible visual generation tool that turns text prompts into live AI sketches using Stable Diffusion! 💡 This tool listens to natural language, generates an image with StableDiffusion, and draws it in real-time using Tkinter and OpenCV — it’s like watching Claude sketch live with every prompt.

SketchStorm – Real-time AI Sketching from Pro

A deep learning application for analyzing MRI brain scans to predict Alzheimer's disease stages using computer vision and explainable AI techniques. https://github.com/IamAbhinav01/Alzheimer-s-MRI-Analysis-with-Deep-Learning.git

Alzheimer's MRI Analysis

ImageGPT is an AI-powered tool that allows users to upload images and ask questions about their content. By combining OCR, image preprocessing, and large language models, it provides structured, accurate, and easy-to-understand answers — all in real time. The system is designed to understand images like documents, screenshots, charts, or handwritten notes and produce meaningful insights, summaries, or explanations. Key Features: OCR & Preprocessing: Extracts text from images using Tesseract OCR with noise removal, thresholding, and grayscale conversion. Text Structuring: Automatically cleans and formats extracted text for better readability. Multimodal Understanding: Combines the uploaded image and user prompt using LangChain Groq to produce context-aware answers. Interactive Interface: FastAPI backend with a dynamic chat frontend, supporting image upload and instant responses. Reusable Pipeline: Modular design allows easy expansion for additional document types or AI capabilities.

ImageGPT: Multimodal AI Assistant

🔹 Built a fully offline, blazing-fast AI coding assistant named Derick that integrates into VS Code using a custom extension and serves responses locally through Ollama. 🔹 Powered by open-source LLMs like StarCoder and CodeLlama, Derick provides intelligent code completions, explanations, and inline help – without relying on cloud APIs. 🔹 FastAPI backend handles prompt routing, model communication, and live logging. 🔹 The extension was built with TypeScript, Yeoman, and Node.js, supporting slash commands and future inline completions.

Derick – Your Local AI Coding Assistant for V

A deep learning application for analyzing MRI scans and detecting different types of brain tumors using PyTorch and Streamlit. The application includes Grad-CAM visualization to highlight the regions that influenced the model's decision.

MRI Tumor Analysis with Grad-CAM

Analyze food images for detailed nutrition info Get personalized recipe suggestions based on what you have Chat with an AI assistant for step-by-step cooking guidance Explore a rich collection of healthy Indian recipes

KitchenElite

your smart assistant for generating quizzes, study prompts, and exam prep content in seconds!

EDU-PROMPTAI

Github Link: https://github.com/IamAbhinav01/Rag_basic_app/tree/main Developed a web-based application that allows users to upload a PDF and ask questions based on its content. The system retrieves relevant document chunks and generates answers using a retrieval-augmented generation (RAG) pipeline. Key Features: PDF parsing and chunking Vector embedding using Cohere Semantic retrieval via FAISS Context-aware response generation with LangChain and Cohere LLM Deployed with Streamlit Cloud

PDF Question Answering App using RAG

Developed an image classification model using Convolutional Neural Networks (CNN) to accurately distinguish between cat and dog images. Achieved 88% training accuracy and 80% validation accuracy over 25 epochs. Built an interactive Streamlit web app allowing users to upload images and get predictions in real time. Tools used: TensorFlow, Keras, Python, Streamlit.

Cat vs Dog Image Classifier using CNN

Developed an AI-powered music generation platform that transforms textual prompts into original musical compositions. Leveraged modern AI techniques to create melody, harmony, rhythm, and style dynamically. Key Features & Technologies: AI Music Generation: Utilized LangChain with ChatGroq for LLM-driven music composition. Musical Parsing & Synthesis: Employed music21 for note parsing and synthesizer for waveform synthesis. Audio Output: Generated high-quality WAV files using scipy.io.wavfile. Backend Integration: Connected a FastAPI backend to handle prompt inputs and music generation requests. Deployment: Dockerized the entire application for cross-platform deployment and easy scalability. Impact / Outcome: Enabled users to generate custom music in real-time from textual prompts. Streamlined AI music creation workflows, reducing manual composition effort. Demonstrated integration of LLMs with procedural music synthesis in a scalable system.

GenAI MUSIC

A smart interview simulator that analyzes resumes, extracts key skills, and conducts real-time mock interviews using LLMs. ✅ Key Features: 📄 Upload resume (PDF) 🧠 LangChain LLM parses skills dynamically — no hardcoding ❓ Scenario-based technical questions generated in real time 🗣️ Supports text and voice answers with live transcription 📸 Snapshot capture simulates a real interview environment 🧠 Instant LLM feedback for each answer 🔧 Built with LangChain, Mistral API, FastAPI, Tailwind CSS 🐳 Fully Dockerized | 📦 Future-ready for MLOps deployment

AI Mock Interviewer  — Resume-Based Real-Time

Campus Copilot is a smart AI-powered assistant designed to help students manage college life efficiently. It provides an intuitive interface to: 📅 Organize and visualize class schedules, events, and deadlines 🔍 Search and filter events by category or keywords 🤖 Interact with a smart assistant powered by LangChain and Groq for natural-language queries ✉️ Send email reminders and agenda updates via SendGrid 🎨 Features a clean Streamlit UI with calendar view and Lottie animation integration Built using: Python, LangChain, Groq API, AstraDB, SendGrid, Streamlit ⚙️ Features Event creation with time, description, and category Filter by event type (Class, Exam, Club, etc.) Email notifications with upcoming agenda Visual calendar (via streamlit-calendar) Embedded AI assistant for Q&A on upcoming tasks

Campus Copilot – Your Personal College Assist

Built a smart, Streamlit-based PDF assistant that enables interactive Q&A over uploaded documents. Integrated LangChain with Groq LLM, HuggingFace embeddings, and FAISS for retrieval. Added session-based memory using RunnableWithMessageHistory and LangSmith for traceable debugging and run tracking. Supports dynamic domain prompting and chat history replay.

DocAI – Domain-Specific PDF Chat Assistant (A

LexAI Justice is a voice-enabled legal assistant that provides instant, conversational answers to Indian law queries. It leverages Retrieval-Augmented Generation (RAG) with vector embeddings stored in AstraDB to answer real-world legal questions with clarity, citations, and voice output. 🔹 Key Contributions: Built an end-to-end Streamlit application with an elegant, professional UI. Used Whisper ASR for speech-to-text transcription, optimized for short queries. Integrated Coqui XTTS for dynamic voice-based responses with speaker cloning. Embedded over 100+ Indian legal documents and stored vector representations in Cassandra (AstraDB) for scalable retrieval. Implemented LangChain’s ConversationalRetrievalChain to enable context-aware question answering.