Aryan Ahuja

Undergraduate · Computer Science & Artificial Intelligence · University of Pennsylvania

Research

Interests: mechanistic interpretability, multilingual NLP, AI safety.

Bridging the Literacy Gap in Government Communication
IEEE iSAI-NLP 2025 (oral) · first author · IEEE Xplore
Fine-tuned T5/BART to translate U.S. government text into plain language with a self-built 20k-pair corpus, a staged training curriculum, and custom constraint losses (+7.1 SARI).
A Meta-Analysis on Image-Based PDF Data Extraction
SSRN, 2025 · first author · SSRN

Projects

Tide
Live · GitHub
A hyper-local fishing-intelligence agent that returns cited, condition-aware recommendations at sub-2s p50 latency. A four-node LangGraph pipeline (GPT-4o-mini handing off to Claude Sonnet) sits over a hybrid Qdrant retriever and five per-species XGBoost models trained on real-time NOAA data.
LangGraph · Qdrant · XGBoost · TimescaleDB · FastAPI · GCP
Visual Delta Highlight Reels
GitHub · Writeup
Automated highlight detection for long body-cam footage, built in a couple of hours. It embeds every frame with CLIP ViT-L/14, takes cosine deltas between consecutive frames to score visual change, finds the salient moments with a rolling-MAD peak detector, and cuts duration-budgeted clips around them.
PyTorch · CLIP · NumPy · OpenCV

Experience

Algoverse AI, Machine Learning Engineer Intern
Nov 2025 – present
Tokenizer-alignment training in PyTorch + Hugging Face across five open-source model pairs (Pythia aligned to Qwen, LLaMA, Gemma, and OPT) on multi-GPU cloud; built an anchor-relative cosine-projection transform and a reproducible evaluation harness.
Infologitech, Software Engineering Intern
Jun – Sep 2025
OCR PDF-extraction pipeline (Camelot / PyMuPDF / Tesseract + GPT-4) at 95%+ accuracy, cutting B2B quotation time 64%; NLP resume matcher screening 2,000+ resumes at 20× manual speed.

Skills

Languages & ML: Python, TypeScript, SQL, LaTeX, PyTorch, Hugging Face, XGBoost, scikit-learn
Frameworks: FastAPI, LangGraph, LangChain, Next.js, React, OpenAI / Anthropic APIs
Data & Infra: PostgreSQL / TimescaleDB, Redis, Qdrant, Apache Kafka, Docker, Terraform, GCP

Food

young me in a chef coat with a plate of lamb chops I made for my family

I've been cooking since I was a kid; that's me at home after a Future Chefs challenge, with some lamb chops I made my family for dinner. It's the same loop I like in research: a hypothesis, a fast iteration, a taste-test.

Contact

Email is hidden from scrapers; it. Also on GitHub and LinkedIn.