Ronak Patel

Assistant: Workflow Automation with AI Medical Summarizer Project Using Gemma Text-to-SQL

• Designed and developed an AI-powered Assistant using Python, Flask, Llama 2, and Ollama, automating blox form creation processes with secure APIs and advanced prompting models.
• Engineered a chatbot interface for intuitive user interactions, dynamically identifying missing fields and streamlining data input workflows.
• Delivered a robust solution that enhanced productivity and significantly improved workflow automation within the application.
• Built a scalable medical summarization pipeline leveraging Gemma model, Flask APIs, and asynchronous programming for efficient, real-time patient summary generation.
• Cleaned and normalized medical data using Pandas, enabling precise summaries with minimal task-specific training through few-shot prompting.
• Optimized the system for large-scale processing, improving throughput and integration with healthcare systems for better data management.
• Developed a Text-to-SQL system using OpenAI, LangChain, and Document DB, enabling seamless query generation from natural language inputs.
• Implemented embeddings-based retrieval to dynamically map user queries to relevant database schemas, enhancing accuracy.
• Delivered an efficient pipeline for generating SQL queries, streamlining database interactions and improving decision-making workflows.

• Developed a robust policy data extraction pipeline utilizing various transformer models, overcoming the 512-token limit in LayoutLMv3 by fine-tuning its tokenizer algorithm.
• Trained an object detection model for Certificate of Insurance (COI) documents, automating the extraction of tabular information and exporting it into CSV files using Python.
• Conducted model training on BERT to efficiently identify and extract important pages from policy PDFs, enhancing data retrieval accuracy and efficiency.

• Invoice data extraction using Deep Learning
• Generating MCQ's using Generative models
• PCB fault detection
• Data Analytics for Recommendation System
• People counter using yolov5 on jetson nano
• Automated the extracted data from documents into Tally Prime software using python

• Policy data extraction using various transformer models
• Solved 512 token limit issue in LayoutLMv3 model by finetuning it's tokenizer algorithm
• Trained object-detection model for COI certificate and then, extracting that tabular information to csv file using python script
• Done model training on BERT model for extracting important pages from policy pdfs