Currently an Extern @PfizerBuilding skills in data analytics, AI-powered data extraction, and transforming complex information into actionable insights through my experience at Pfizer.
I am currently engaged in projects focused on AI and data extraction, aiming to enhance my skills and make an impact in the tech field.
.jpg)
Pfizer · ✅ Verified by Extern · ⏱️ In progress
Prototype AI-powered document intelligence with Pfizer—using OCR, LLMs, and RAG to automate real enterprise PDF workflows and build a standout portfolio project.
Building skills in data analytics, AI-powered data extraction, and transforming complex information into actionable insights through my experience at Pfizer.
I’m an aspiring Data & AI professional passionate about using technology and data to solve problems and uncover meaningful insights. Through my experiences at Pfizer and UMass Lowell, I’ve been developing hands-on skills in AI-powered data extraction, data analytics, and problem-solving. I enjoy learning new technologies, working with data, and exploring how AI can transform complex information into actionable insights. I’m excited to continue growing and connect with professionals who share an interest in data, technology, and innovation.
Externships
Pfizer Advanced: AI-Powered Document Insights & Data Extraction Externship
Pfizer
Experience
Information Analyst Intern
Mosaic Lowell · May 2024 - Aug 2024
Education
University of Massachusetts at Lowell
Bachelors of Science in Business Administration · Class of 2027
Skills
✅ Verified by Extern · ⏱️ In progress
Prototype AI-powered document intelligence with Pfizer—using OCR, LLMs, and RAG to automate real enterprise PDF workflows and build a standout portfolio project.
The work prototyped an AI-driven pipeline that converted scanned pharmaceutical PDFs into machine-readable outputs. Deliverables included Colab notebooks for data cleaning and image preprocessing, PyMuPDF word-level extracts with bounding boxes and accuracy notes, LLM tokenization documentation, a comparative OCR analysis that favored PaddleOCR for layout preservation, and a working RAG pipeline
I produced runnable Colab notebooks and analyses: LLM tokenization documentation, cleaned structured data and preprocessed images, PyMuPDF word-level extracts with bounding boxes and validation recommendations, a comparative OCR analysis identifying PaddleOCR as best at preserving layout, and a working RAG pipeline with hybrid retrieval.
The submission explained LLM tokenization and probability-based token selection, described how models were trained on large datasets, and used ChatGPT and Grok as examples to illustrate how responses are generated.
I worked in Google Colab to clean structured data with Pandas, fix encoding and text formatting, and preprocess scanned images with OpenCV and PIL. The submission showed cleaned text files, converted numeric types, and a processed image ready for OCR.
I processed multi-page SDFs with PyMuPDF to extract word-level text and bounding boxes, evaluated overall accuracy, noted challenges reconstructing tables and lines, and recommended regex validation and label-based grouping to improve date and field extraction.
I evaluated Tesseract, EasyOCR, and PaddleOCR on a scanned pharmaceutical certificate, ran preprocessing, compared outputs for key-value pairs, tables, and noise resistance, and produced a comparative analysis showing PaddleOCR preserved layout and key-value structure best.
During the externship I built and optimized a RAG pipeline. I chunked documents (512 token chunks, 50 overlap), used all-MiniLM-L6-v2 embeddings, and implemented hybrid retrieval (vector search plus BM25). The system returned concise, document-grounded answers to sample prompts.