Portfolio - Abhinav Painuli

Hey! I am Abhinav Painuli and this portfolio represents all my work in Data Science.

🧑‍💼Internships

• Machine Learning Intern – IIT Bombay (Oct 2024 – Mar 2025, Remote)
  • Fine-tuned multiple LLMs from Hugging Face for optimized performance on custom datasets.
  • Developed multilingual ASR pipelines using IndicWav2Vec, IndicConformer, and IndicWhisper models with diarization, Voice Activity Detection (VAD), and XML generation.
  • Built ASR error detection pipelines to correct segmentation errors, number mismatches, compound words, and underrepresented characters.
  • Conducted no-shot, one-shot, and few-shot inference with ChatGPT to evaluate hypothesis accuracy.
  • Generated embeddings for analyzing speakers and transcriptions across audio datasets.

🕵️👩‍⚖️Agentic AI

• Self-Improving FloodHK Agent Swarm
  • Built a self-improving 7-agent LLM disaster intelligence swarm for real-time flood risk prediction across Hong Kong
• PeaceProject
  • Integrated live news, geopolitical tensions in the agentic swarm based prediction.
• Automated WhatsApp Agent LifeLift Coach
  • Built a smart Life Coach chatbot assistant with n8n, GPT-4.1 mini and Gemini 2.0 Flash-Lite.

✉️Natural Language Processing

• SemRAG: Semantic Knowledge-Augmented RAG for Improved Question-Answering

  • Built a full SemRAG system combining semantic chunking, vector + keyword hybrid search, and KG-based reasoning.

• Conversational Document Assistant using RAG

  • Built a RAG-based chatbot on pdf documents with Zephyr-7b, HuggingFace and ChromaDB.

• Vector Database Retriever using Weaviate, HuggingFace, and FastAPI

  • Built a vector database retriever using Weaviate and HuggingFace API and deployed it using FastAPI

• TutKids ChatBot using Google PaLM and FastAPI

  • Implemented a chatbot using Google PaLM and built an api using FastAPI.

• Table-based Basic Question-Answering with Tapas

  • Implemented a basic question-answerer from any database uploaded using transformers deployed on streamlit.

• Basic Speech Recognizer and Sentiment Analyser

  • Implemented a speech recognizer and sentiment classifier on the converted text using nltk.

• Spam Mail Predicion

  • Implemented a basic spam mail prediction using Logistic Regression.

🔎Data Analysis

• Customer Analysis by Clustering
  • Analysed customer behaviour by clustering them using K-Means.

⚙️Recommendation Systems

• Book Recommender Website using Streamlit and Docker
  • Deployed a website for recommendations of books using streamlit and docker.

🖼️Image Processing

• Analysis of Anatomical Tracings of Lesions After Stroke

  • Segmentation on Brain MRI scans for the tracing of lesion dute to stroke.

• Self-Driving Car

  • Implementing an autonomous self-driving car using CNN and Udacity car simulator.

• ResNet-50 Implementation for Image Classification

  • Implementing ResnNet-50 for image classification of 15 sport balls by transfer learning.

• Custom Face Dataset StyleGan-3

  • Implementing StyleGan-3 on custom dataset for high-resolution images.

• CIFAR-10 RNN

  • Classfication of cifar-10 dataset using RNN.

• Handwritten Digit CLassification ANN

  • Classification of handwritten digits using ANN.

🎵Audio Processing

• Hindi Whisper Automatic Speech Recognition

  • Automatic transcription of Hindi audiofiles using Whisper OpenAI and Whisper-medium model fine-tuned for Hindi.

• Speech MNIST Classification

  • Classifying digits from 0 to 9 using audio data.

• Audio Data Analysis and Creation

  • Analysis of music audio data using spectrograms and converting it back to audio.

👀Computer Vision

• RPS Predictor Bot using Markov Chains

  • RPS predictor bot using Markov chains with the ability to understand patterns.

• YOLO Vehicle Detection

  • Vehicle detection using YOLOv8 by transfer learning.

• Deployed Face Detection Attendance System

  • Face attendance system with the functions of taking, reading and compiling attendance deployed using streamlit.

• Selfie Smile Detection and Capture

  • Automatically captures a photo if smile detected in the selfie camera.

• Real-time-OCR-Camera

  • OCR using a webcam to detect text in real-time.

🎨Generative Adversarial Networks (GANs)

All of these projects required image processing as well.

• Anime Face Generator DCGAN

  • Generating anime faces using DCGAN without feature-selection.

• Lego Image Generator GAN

  • Generation of middish LEGO images using DCGAN.

• MNIST Image Generation and Noise Reduction Autoencoder

  • Generating MNIST images and cleaning noise from the images using Autoencoder.

• Cat Image Generator LSGAN

  • A cat image generator made with the help of official pytorch documentation.

• Celeba Face Generator DCGAN

  • Generation of face images using DCGAN without feature-selection.

🏆Kaggle Competitions

• Spaceship Titanic Prediction
  • Classification of transportation of people in the Spaceship titanic using XGBoost and RandomForest classifier.

📚Books and Tutorials

Some of the codes I have tried to reproduce from ML and DL books and tutorials.

• Hands-On Machine Learning with Scikit-Learn and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems

• Zero to Deep Learning

• Siddhardhan Tutorials

• Deep Learning Specialization Andrew NG

• MIT 6.S191: Deep Generative Modeling

📄Research Papers

Research papers I haved tried to reproduce.

• SemRAG: Semantic Knowledge-Augmented RAG for Improved Question-Answering

• Analysis of Anatomical Tracings of Lesions After Stroke

• Euler State Networks: Non-dissipative Reservoir Computing

• Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks