I build models and pipelines that extract structured data and surface insights from documents, developed during a Pfizer externship.
Portfolio of externship work showing models, extraction pipelines, and summaries built to convert document content into structured outputs and insights.
.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.
I build models and pipelines that extract structured data and surface insights from documents, developed during a Pfizer externship.
I am Changqi Sun. I completed the Pfizer Advanced: AI-Powered Document Insights & Data Extraction externship, working on AI methods for extracting and summarizing information from documents and turning document content into structured data.
Externships
Pfizer Advanced: AI-Powered Document Insights & Data Extraction Externship
In progress
Education
western university
mechanical and materials engineering · Class of 2028
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 an AI document-insights pipeline combining OCR, large language models, and retrieval-augmented generation to process enterprise PDFs. The work reviewed foundational LLM concepts and implemented components of a document-intelligence workflow. Deliverables included research notes and technical explanations of how LLMs operate.
I produced a technical summary that explained LLM tokenization, Transformer attention mechanics, sequential token prediction, plus descriptions of pre-training, fine-tuning, and human-feedback training steps.
The submission summarized how LLMs tokenized text, used Transformer attention to weigh context, and predicted tokens sequentially. It described pre-training on large datasets and subsequent fine-tuning and human feedback to improve instruction following.
The module required cleaning structured and unstructured data and improving OCR inputs. I cleaned tabular and JSON data with pandas, implemented text-standardization steps, and applied grayscale, denoising, CLAHE, and adaptive thresholding to produce a binarized image for OCR.
