Hi, I'm Aryan Agrawal. I'm passionate about building reliable, scalable software, and increasingly, intelligent systems that do real work. I recently graduated with my Master's in Computer Science from Indiana University Bloomington (Spring 2026), where I worked as a Graduate Research Assistant at the Kelley School of Business, and I'm an ex-SWE Intern at Gametime Hero. I'm also an AWS Certified Cloud Practitioner, which has been a nice complement to the cloud work I do across my projects.
I enjoy shipping production-quality features, building clean APIs and workflows, and turning messy real-world problems into systems that are easy to trust and maintain.
I believe every line of code should create value, improving reliability, saving time, or making complex workflows simpler for users.
My most recent deep dive has been into how retrieval-augmented systems actually behave once you stop taking "it works" at face value and start measuring it. SourceLens started as a simple "ask questions about your PDFs" app, and turned into something much more interesting once I started asking how do I know this is actually good?
Along the way, I found a real, counterintuitive result: a retrieval upgrade that made the system objectively more precise also quietly broke it for a specific class of questions, ones that need context to be pieced together, not just looked up. Finding that, understanding why it happened, and fixing it without losing the original improvement ended up being the most satisfying part of the whole project.
If you're curious how a RAG pipeline can look great on one metric and be worse for real users at the same time, this one's worth a look.
Explore the project here: https://github.com/aryanag7/SourceLens-RAG-Application
My first industry internship = tons of learning beyond just syntax. I got comfortable in a big codebase, wrote small, clean modules, and learned how good pull requests + feedback loops keep teams fast. Team ritual: demo-o'clock, show a tiny win, snag quick feedback, and ship it. Also mastered the ancient art of turning "works on my machine" into "works on everyone's machine." 😄
My big win: I owned and shipped a reusable translation/i18n component, enabling the app to support multiple languages consistently. One place to manage copy, instant language switching in the UI, and an easy path for new pages to "just work" with translations, less duplicate effort, more consistency, and future languages made simple.
To keep things sturdy, I improved reliability through testing + CI/CD and helped fix production issues through debugging and log-based investigation. I also wired organiser sign-ups into ClickUp (auto-tasks with context). I cleaned up issues that caused stale screens or duplicate actions, fewer hard refreshes, smoother flows, and happier users.
Tech I enjoyed: Angular + TypeScript (Signals/RxJS), Firebase, Git/GitHub, ClickUp.
During my Master's, I did research at the Kelley School of Business focused on building data pipelines and applying ML/NLP to messy real- world datasets. I extracted and processed on-chain NFT data (queried via GraphQL) and associated media (IPFS), then ran structured analysis and modeling on top of curated datasets to understand pricing behaviour. I also built ETL/NLP workflows over large text corpora (Reddit post-comment hierarchies, financial news) and used statistical methods to surface patterns and relationships between sentiment and market behaviour.
I've been going deeper into LLM-powered systems and agentic workflows, and how to build automations (and evaluate them properly) so they're accurate, auditable, and safe in production. I'm especially interested in:
- Retrieval and vector search (RAG-style patterns), and how to actually measure whether they're working
- LLM workflows and agents (tool use, decision routing, human-in-the-loop)
- Building reliable automation with validation, deduplication, and logging
Upload PDFs, ask questions, get answers grounded in the actual source material, not guesses. What started as a straightforward RAG build turned into a real investigation once I started rigorously evaluating it: a retrieval upgrade improved precision but quietly broke answers for a whole category of questions, and tracking down why (and fixing it without losing the improvement) taught me more than the initial build did.
Explore the project here: https://github.com/aryanag7/SourceLens-RAG-Application
I built an LLM + OCR automation agent to streamline a real front-desk workflow. It turns a manual, error-prone process into a consistent, auditable pipeline.
What it does:
- Parses shipping labels using OCR and an LLM to extract recipient info
- Resolves recipients using a Google Sheets directory (faculty + PhD students)
- Sends the email automatically when confidence is high, otherwise drafts for review
- Tracks status and prevents duplicates using idempotent tracking-ID dedup and logging
Impact:
- Reduced per-label manual effort by ~85%
- Saved ~30+ hours/month and improved consistency for notifications
Explore the project here: https://github.com/aryanag7/LLM-Powered-Package-Notification-Agent
Built with a teammate as part of a course project, this serverless AWS chatbot uses Lex, Lambda, and SageMaker to provide cost insights and EC2 rightsizing recommendations. It taught me how to design reliable cloud workflows with async processing and scalable services.
With 100+ queries processed and ~97% Lambda success rate, I learned how serverless systems can deliver real value when built with reliability in mind.
Explore the project here: https://github.com/aryanag7/AI_ChatBot_Resource_Utilization
I built a full-stack homestay platform to simplify the booking experience for hosts and guests. Using Node.js, Express.js, and MongoDB, I implemented secure authentication, image uploads with Cloudinary, and Mapbox for interactive maps.
Revisiting it recently, I also tracked down and fixed a real performance bottleneck in how listings were being loaded, cutting response times nearly in half. A good reminder that "done" projects are still worth going back to with a more experienced eye.
Explore the project here: https://github.com/aryanag7/HomeStay
I built a language-agnostic data visualization platform that lets users transform data using Python or R and generate compelling visualizations without being tied to one language.
It supports interactive charts (HTML), static exports (PNG), and 3D visualizations, with real-time preview, download options, and a responsive UI built with Angular and Angular Material.
Explore the project here: https://github.com/aryanag7/Language-Agnostic-Visualization
I like work where reliability, scale, and clarity matter. I enjoy:
- Building and improving user-facing features
- Designing APIs and backend services
- Debugging and improving system behavior in production
- Automating messy workflows with careful validation + fallbacks
- Exploring ML/LLM systems in a practical, product-focused way, and actually measuring whether they work
Currently, I'm focusing on:
- Building more agentic workflows (tool use, routing, evaluations)
- Going deeper on retrieval evaluation, not just building RAG systems, but rigorously measuring them
- Improving my full-stack depth with React/TypeScript and stronger backend systems
I'm always happy to connect with people building in software, cloud, and applied AI.
- 📧 Email: aryanagrawal2310@gmail.com
- 🌐 LinkedIn: https://www.linkedin.com/in/agrawal-aryan23/
- 📄 Resume: https://drive.google.com/file/d/1j5dyoDpsS--tYLRGFhAvWqZJwzpsTVfg/view?usp=sharing
When I'm not coding, you'll find me:
- Watching Test cricket, my favourite sport to follow
- Enjoying movies and TV series, huge fan of binge-watching
- Playing football, cricket, badminton, and pickleball
- Spending time reflecting, learning, and exploring new ideas