Currently an Extern @PfizerI'm an aspiring database architect passionate about SQL and building connections in technology.
Discover my work in SQL and database architecture, showcasing my learning and growth during my externship.
.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.
Quantified $246K annual revenue impact by conducting two-proportion z-tests and 10,000-iteration bootstrap resampling across 90,189 players, identifying a 4.6-point Day 7 retention decline among high-value users. Built R Markdown dashboards visualizing segmentation analysis and statistical confidence intervals, delivering a campaign effectiveness recommendation grounded in binomial distribution methodology to non-technical stakeholders.
I'm an aspiring database architect passionate about SQL and building connections in technology.
I'm an aspiring professional currently completing an externship in SQL and Database Architecture with Breaking Games. With a Bachelor’s degree, I'm eager to grow my skills and build valuable relationships in the industry.
Externships
Pfizer Advanced: AI-Powered Document Insights & Data Extraction Externship
Pfizer
Experience
AI Document Intelligence Extern
Pfizer · Jul 2026 - Present | Remote
Data Scientist AI Training Expert
micro1 · Jun 2026 - Present | Remote
Data Analyst Intern
Sallie Mae, Chief Data Office · May 2025 - May 2026 | Remote
Education
New York University — New York, NY
B.A. Data Science, Business Studies | GPA: 3.6 · Class of 2026
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 externship prototyped AI document-intelligence methods for pharmaceutical PDFs, combining OCR, computer vision, NLP, and LLMs. Deliverables included a concise technical explanation of LLM mechanics, Python notebooks that cleaned and standardized messy documents, and image-denoising and contrast tests to improve OCR input. The work also produced structured JSON exports from SDF PDFs using
I produced a concise explanation of how LLMs predict tokens sequentially and delivered Python notebooks plus extraction scripts that cleaned text, tested image-denoising and contrast for OCR, and exported structured JSON from SDF PDFs.
The module required explaining LLMs. I described that models learned statistical patterns from large text corpora, predicted the next token sequentially, and produced responses by sampling those predictions. The submission was a concise explanation of LLM mechanics.
The project inspected messy documents, wrote Python notebooks to clean and standardize text (fixed encodings, normalized prices, removed tags/emoji, extracted contacts) and tested image denoising and contrast steps to improve OCR input.
I extracted text and bounding boxes from SDF PDFs, applied regex and anchor-phrase logic, and clustered table rows to identify dates and vendor fields. The submission produced structured JSON and a shared PDF/drive deliverable.
A scanned certificate had rotated sidebar text and handwritten signature noise. I ran Tesseract, PaddleOCR, and EasyOCR on the same 300 DPI PNG, inspected raw text and bounding boxes, debugged PaddleOCR installation and API mismatches, and produced a comparison showing PaddleOCR delivered the most…
I ran end-to-end RAG notebooks that parsed PDFs, applied SentenceSplitter chunking, generated embeddings, and compared BM25, vector, and hybrid retrieval. I debugged indexing order and scoring-scale differences, and produced a reproducible Colab pipeline with findings about single-page limits.
Quantified $246K annual revenue impact by conducting two-proportion z-tests and 10,000-iteration bootstrap resampling across 90,189 players, identifying a 4.6-point Day 7 retention decline among high-value users. Built R Markdown dashboards visualizing segmentation analysis and statistical confidence intervals, delivering a campaign effectiveness recommendation grounded in binomial distribution methodology to non-technical stakeholders.
Surfaced $17.7K revenue recovery opportunity by analyzing 270,000 users in BigQuery, identifying 93.6% churn and building cohort retention models to reveal a critical Month 1 player re-engagement window. Created Tableau dashboards tracking engagement metrics, purchase patterns, average revenue per user, and channel attribution, delivering actionable VIP outreach insights to cross-functional stakeholders.
