Amazon Externship via ExternI have practical work experience and I’m exploring the next steps in my career, sharing projects that show what I can do.
Project writeups, prototypes, and outcomes that reflect the skills I’ve practiced so far and the directions I’m exploring next.
Built a privacy-first, on-device AI reading assistant for research papers, increasing reader comprehension accuracy by 25%. Delivered contextual explanations in under 4 seconds using quantized Gemma models and an 8,192-token context window.
Engineered an AI-powered scheduling platform that syncs Google Calendar to assign employees to tasks by skill and availability. Achieved ∼80% recommendation accuracy against 30 labeled shift assignments while supporting 50+ employees.
I have practical work experience and I’m exploring the next steps in my career, sharing projects that show what I can do.
I’m Rayyan Rayyan. I have some hands-on work experience and I’m exploring where to take my career next. I build and learn through projects, and I present my work here to show what I can do and where I want to grow.
Externships
Amazon Operational Strategy & People Analytics Externship
In progress
Experience
Research Intern | Python, TrackPy, NumPy, Pandas, ImageJ/Fiji
Kandula Research Group | Manasa Kandula (PhD) · October 2025 – Present
Machine Learning Research Intern | Poster | Python, PyTorch, MNE, scikit-learn
UMass Amherst | Sarmistha Sarna Gomasta (PhD) · July 2026 – September 2026
Software Engineer Intern | Python, LM Studio, FFmpeg, MLX Whisper
Myco | Web3 Video Streaming & Entertainment Platform · May 2026 – July 2026
Full Stack Developer Intern | React, TypeScript, Express, Firebase, MongoDB
Astrik Digital | Software Development Agency · January 2025 – April 2025
Education
University of Massachusetts Amherst
B.S. in Computer Science, Commonwealth Honors College | GPA: 3.95/4.00 · Class of 2028
Skills
Built a privacy-first, on-device AI reading assistant for research papers, increasing reader comprehension accuracy by 25%. Delivered contextual explanations in under 4 seconds using quantized Gemma models and an 8,192-token context window.
Engineered an AI-powered scheduling platform that syncs Google Calendar to assign employees to tasks by skill and availability. Achieved ∼80% recommendation accuracy against 30 labeled shift assignments while supporting 50+ employees.
Developed a responsive platform enabling UMass students to discover and explore more than 50 campus clubs and organizations. Implemented tag-based recommendations, persistent saved clubs, and filtering by category, meeting time, and commitment level.
✅ Verified by Extern · ⏱️ In progress
Imagine shaping how AMAZON — yes, THE Amazon — welcomes and keeps its workforce. Every year, Amazon loses billions because new hires leave before Day 90. You’re going to fix that. In this externship, you’ll web scrape real-world employee reviews from across the web, extract actionable insights, and design persona-driven strategies to help Learning Ambassadors connect smarter, faster, and more humanely. It’s part research, part strategy—and all about learning how people data drives big decisions.
The externship analyzed employee feedback and operational context for Amazon fulfillment centers to surface patterns affecting early-tenure retention. I combined scraped employee reviews with role and process analysis, then synthesized persona-driven findings and recommended interventions. The work produced a set of evidence-backed observations and a persona framework.
.png)
I produced a CSV-style workforce-role matrix for an Amazon fulfillment center that listed each role, main responsibilities, frequent collaborators, common tools, and notes tying roles to operational impact.
I compiled a workforce-role matrix for an AFC, listing each role, main responsibilities, who they worked closely with, common tools, and notes. The deliverable produced a clear CSV-style table linking roles to operational impact and systems used.
I reviewed Glassdoor reviews, selected six examples (three positive, three negative), annotated each with why it was strong, moderate, or weak, and noted concrete issues or drivers that could inform action.
The project cleaned Glassdoor and YouTube review text, kept four cleaned columns alongside originals to preserve context, noted stopword removal issues (negation loss), and produced a shared cleaned dataset for keyword and sentiment analysis.