Pfizer Externship via Extern
Pfizer Externship via ExternI’m a Computer Engineering student at Iowa State building OCR and RAG prototypes, Android apps, and embedded systems—turning complex requirements into practical, tested tools.
Selected work across AI document intelligence, Android lab workflows, autonomous robotics, and RISC-V processor design—built with a focus on integration, validation, and usable interfaces.
College Project · May 2026
A native Android application for lab workflows, built with Java and integration-ready interfaces for a Spring Boot/MySQL backend. I led frontend design and implementation, creating the visual system, reusable screens, and navigation.
Built a TM4C123-based mobile robot for autonomous navigation and object detection (November–December 2025). The system combines embedded C control logic, sensor integration, and a Python interface for live visualization and click-to-move control.
Designed a single-cycle RISC-V processor in structural VHDL during summer 2026. The datapath and control path support arithmetic, load/store, branch, and jump instructions.
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’m a Computer Engineering student at Iowa State building OCR and RAG prototypes, Android apps, and embedded systems—turning complex requirements into practical, tested tools.
I build across software and hardware, from Django web features to embedded C robotics and structural VHDL processors. As LabFlow’s frontend design lead on a four-person team, I helped deliver an Android lab-workflow app that ranked 1st in section and 2nd overall among 40+ teams; testing across three lab types measured up to 60% less lab time. I completed two years of Computer Science and Engineering coursework at Nirma University before transferring to Iowa State. Outside engineering, I support live athletics events and have volunteered in elder care.
Externships
Pfizer Advanced: AI-Powered Document Insights & Data Extraction Externship
In progress
Experience
Production Crew
Iowa State University Athletics · Apr 2026
Web Development Intern
Relax India Group / Varlyq Technologies · May 2024 - Jul 2024
Education
Iowa State University, College of Engineering
B.S. in Computer Engineering (expected December 2027) · Class of 2027
Nirma University
Computer Science and Engineering coursework, 2+2 transfer program (2023–2025) · Class of 2025
Skills
College Project · May 2026
A native Android application for lab workflows, built with Java and integration-ready interfaces for a Spring Boot/MySQL backend. I led frontend design and implementation, creating the visual system, reusable screens, and navigation.
A native Android application for lab workflows, built with Java and integration-ready interfaces for a Spring Boot/MySQL backend. I led frontend design and implementation, creating the visual system, reusable screens, and navigation.
Led frontend delivery within a team of two frontend and two backend developers. Built reusable interfaces and coordinated backend integration, resolving major integration issues before the final demo.
Testing across three lab types measured up to 60% lab time reduction. LabFlow ranked 1st in section and 2nd overall among more than 40 teams.
Designed a single-cycle RISC-V processor in structural VHDL during summer 2026. The datapath and control path support arithmetic, load/store, branch, and jump instructions.
Designed a single-cycle RISC-V processor in structural VHDL during summer 2026. The datapath and control path support arithmetic, load/store, branch, and jump instructions.
Integrated the ALU, control unit, register file, immediate generator, memory path, and PC-selection logic into a single-cycle RV32I design.
✅ 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.
Since August 2026, I have prototyped enterprise PDF workflows using OCR, image processing, LLMs, and retrieval-augmented generation. The work extracts and structures information from scanned and digital documents using PyMuPDF, pdfplumber, Tesseract, PaddleOCR, LlamaIndex, Mistral, Gemini, and Gradio.
.jpg)
Evaluated extracted outputs against source documents using Python and regex-based validation. Documented failure cases and accuracy limitations, and mapped pharmaceutical-documentation requirements into prototype workflows and technical findings.
The submission summarized how large language models processed text: described transformers for context, training on large corpora, next-word prediction, and further improvement via fine-tuning and human feedback.
The module provided messy tabular, JSON, text and image inputs. I cleaned a 5,000-row table to a validated 4,947-row CSV with 53 records set aside for review, flattened JSON orders, standardized text while preserving prices and contacts, and produced denoised image variants for OCR testing.
The submission processed a three-page SDF with PyMuPDF, preserved word-level bounding boxes and page provenance, reconstructed multi-word labels and a six-row test table, parsed five calendar dates and 17 labeled fields, and saved the structured JSON plus an executed notebook.
I processed a scanned Certificate of Quality at 300 dpi, ran Tesseract, PaddleOCR, and EasyOCR, and retained raw text, confidences, coordinates, and annotated images. I compared field matches, reading order, rotated text, and region counts, and reported per-engine errors and timings.
The project parsed a three-page packet with PyMuPDF, applied sentence-aware chunking and MiniLM embeddings, merged BM25 and vector results with reciprocal-rank fusion, reranked candidates with a cross-encoder, and produced Gemini answers with page citations. Retrieval traces and JSON exports were…
I tested three RAG configurations on a three-page SDF PDF, logging retrieved chunks, runtimes, and generated answers. I compared candidate pools and reranker effects, recorded raw IDs and scores, and rated answer completeness, finding Configuration B retrieved the key table row and produced the…
