Pfizer Externship via ExternI’m a passionate learner diving into SQL and database architecture, ready to make an impact.
Explore my work in SQL and database architecture, showcasing my growth and the skills I'm developing.
.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.
Surfaced a funding dependency risk, with the top 10% of donors giving roughly 89% of dollars, for First Tee Triangle, a Raleigh youth development nonprofit, by building a donor-level data model from raw gift records using Python. Identified trends and risks for strategic planning by flagging donors at risk of lapsing, and delivered an end-to-end analysis report with interactive dashboards covering donor retention, concentration risk, and campaign performance.
I’m a passionate learner diving into SQL and database architecture, ready to make an impact.
I'm on the cusp of launching my career with hands-on experience in SQL and database architecture. Currently, I'm enhancing my skills through an externship at Breaking Games, where I’m learning to navigate the complexities of data management.
Externships
Pfizer Advanced: AI-Powered Document Insights & Data Extraction Externship
In progress
Experience
Data Automation and Analytics Intern
Greentree Capital, Chicago, IL · May 2025 6 July 2025
Merchandise Associate
HomeGoods, Apex, NC · August 2023 6 December 2024
Education
North Carolina State University, Raleigh, NC
Bachelor of Science in Statistics, Minor in Biology · 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-driven document intelligence for enterprise PDFs, combining OCR, retrieval-augmented generation, and LLM concepts. Deliverables included technical notes on LLM training and fine-tuning, and runnable Python Colab notebooks for data cleaning, JSON flattening, text standardization, and image preprocessing for OCR pipelines. The work also produced a notebook that
I documented LLM training dynamics and fine-tuning, and I produced Colab notebooks that implement data-structure exercises, Pandas cleaning, JSON flattening, text standardization, image preprocessing for OCR, and a field-extraction pipeline for multi-page SDF PDFs.
The submission described LLM training on large text corpora, the next-word prediction objective, parameter updates on errors, and how models generate responses token by token. It noted fine-tuning with human feedback and that outputs reflect learned patterns, not factual understanding.
The externship notebook collection showed Python exercises and data pipelines. I wrote Colab notebooks that demonstrated data-structure practice, control flow and functions, Pandas cleaning, JSON flattening, text cleaning steps, and image preprocessing workflows; each notebook contained runnable…
The project processed multi-page SDF PDFs with Python, used PyMuPDF to extract text and bounding boxes, applied regex and anchor-phrase logic to locate fields, and produced a runnable Colab notebook showing the extraction pipeline.
Surfaced a funding dependency risk, with the top 10% of donors giving roughly 89% of dollars, for First Tee Triangle, a Raleigh youth development nonprofit, by building a donor-level data model from raw gift records using Python. Identified trends and risks for strategic planning by flagging donors at risk of lapsing, and delivered an end-to-end analysis report with interactive dashboards covering donor retention, concentration risk, and campaign performance.
Prioritized where intervention has the most impact by finding that 2 of 15 care types drive half of diagnosed spend and 86 of 3,918 facilities carry half of all spending, by writing SQL queries with multi-table joins, aggregations, and window functions against five years of claims data ($99.1B billed) in a Snowflake cloud data warehouse. Detected and drove to closure five data defects, including one that overstated a key figure by roughly 543,000x, by investigating discrepancies, documenting root cause and the corrective SQL logic, and validating the fix. Kept 47 AI-assisted analysis queries reliable by specifying the plan and constraints up front, reviewing every output, validating against independent small-sample tests, and running a second AI agent as an additional review pass.
Built an early-warning predictive model in R to flag high-risk cases for intervention before failure; benchmarked Random Forest against Logistic Regression, Lasso, and XGBoost and selected on test PR-AUC of 0.876. Evaluated model performance across gender and race subgroups and tuned decision thresholds within each subgroup to keep predictions equitable across populations.
Identified 81% of at-risk individuals as candidates for early screening outreach by building a threshold-tuned LightGBM model in Python on CDC survey data. Used SHAP to explain key risk drivers such as BMI, blood pressure, and income level, turning model output into factors public health teams can act on.
