Image-Based Product Classification:
● Designed a MobileNetV2-based CNN with TensorFlow to classify retail products from shelf images.
● Used OpenCV for preprocessing and achieved 92% accuracy using transfer learning.
● Deployed the model via FastAPI, collaborated with cross-functional teams for integration, and supported performance monitoring in internal test environments.
Invoice OCR & Field Extraction:
● Built an OCR-based system to extract structured data (Invoice No., Date, Total, Vendor) from semi-structured PDF invoices.
● Used Tesseract OCR with OpenCV preprocessing for improved text clarity.
● Trained a field classifier using TF-IDF and Logistic Regression with 91% extraction accuracy.
● Deployed via FastAPI and integrated with internal finance tools, reducing manual invoice processing by 60%.
Product Expiry Date Detection from Packaging:
● Built an OCR-based system using Tesseract and OpenCV to automatically detect and extract expiry dates from printed labels on packaged goods.
● Applied preprocessing techniques such as grayscale conversion, blurring, and adaptive thresholding to enhance text clarity.
● Used regular expressions to validate extracted dates and flag expired products.
● Achieved high reliability in varied packaging conditions.
● Deployed the solution using FastAPI and integrated it with the warehouse dispatch system to prevent the shipment of expired items.
Projects:
1. FAQ-Based RAG Chatbot.
● Developed a RAG FAQ Chatbot using LangChain, FAISS and ChromaDB for handling FAQ-based queries. Integrated both vector databases to optimize data retrieval and improve chatbot response accuracy.
● Implemented PDF-based document processing, allowing users to upload files and store embeddings for retrieval.
● Utilized Hugging Face embeddings and a Flan-T5 language model for accurate response generation.
● Built a FastAPI backend with endpoints for PDF uploads and query-based retrieval.
● Applied text chunking techniques using RecursiveCharacterTextSplitter to enhance retrieval accuracy.
● Gained hands-on experience in Vector Databases, LLMs, FAISS, and API development.
2. Intent-Based Chatbot.
● Developed a rule-based chatbot to handle predefined user intents such as greetings, product inquiries, and pricing details.
● Implemented intent-response mapping for structured and efficient query handling.
● Used Gradio to build an interactive user interface for real-time chatbot interactions.
● Applied Python and NLP techniques to improve chatbot response accuracy.
● Gained hands-on experience in chatbot development, user interaction design, and Gradio deployment.
3. Image Generation.
● Built a generative model leveraging Stable Diffusion to create high-quality images from textual descriptions.
● Utilized PyTorch and Hugging Face Diffusers library to implement the architecture and manage training workflows.
● Gained expertise in Generative AI, Stable Diffusion Models, and advanced Python programming for model implementation.
4. Object Detection.
● Designed and implemented an object detection system using the YOLO (You Only Look Once) algorithm to identify and classify objects in real-time.
● Used OpenCV and PyTorch to deploy the model for real-world applications, including live video feeds.
● Gained hands-on experience in Computer Vision, Deep Learning, and efficient object detection frameworks.
5. Image Classification.
● Developed an image classification model using MobileNetV2 as a feature extractor.
● Applied data augmentation techniques (rotation, shifting, zooming, flipping) to enhance model generalization.
● Utilized TensorFlow & Keras to train and fine-tune the model for improved accuracy.
● Processed images using ImageDataGenerator for efficient loading and augmentation.
● Evaluated model performance using test datasets and optimized hyperparameters for better results.
● Gained hands-on experience in Transfer Learning, Deep Learning, and
● Image Processing.
6. Movie Recommendation System.
● Developed a content-based recommendation system using a dataset of Telugu movies.
● Utilized TF-IDF (Term Frequency-Inverse Document Frequency) to extract key features from textual data and calculated cosine similarity to find relevant movies.
● Enhanced user experience by providing accurate and personalized movie recommendations based on user preferences.
● Gained hands-on experience in Python, Pandas, and Scikit-learn, with a focus on NLP techniques.
7. NLP Tasks Using Pretrained.
● Developed a comprehensive NLP system to perform tasks such as Sentiment Analysis, Named Entity Recognition (NER), and Text Generation using pretrained Transformer models from the Hugging Face Transformers library.
● Integrated models like BERT for NER and sentiment analysis, and GPT-2 for text generation, leveraging transfer learning for rapid deployment.