Skip to content
View pedromussi1's full-sized avatar

Block or report pedromussi1

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
pedromussi1/README.md

Pedro Mussi

Incoming MS in Artificial Intelligence at the University of South Florida (Fall 2026); B.Sc. in Computer Science, USF (2023).

I build at the intersection of LLM fine-tuning, retrieval-augmented generation, and on-device computer vision.

What I'm working on

  • paper-tldr — QLoRA fine-tune of Llama 3.2 3B Instruct for scientific abstract → TL;DR. ROUGE-1 +13%, ROUGE-2 +27%, ROUGE-L +22%, BERTScore F1 +4.8% over zero-shot baseline on a 618-pair test set.
  • PaperPal — Local RAG over research PDFs with page-anchored citations. Runs entirely offline (Ollama + ChromaDB), with optional Groq cloud fallback. Live demo.
  • ASL-Translator — Real-time browser-only American Sign Language fingerspelling translator. MediaPipe HandLandmarker + ONNX Runtime Web; 98.06% test accuracy on Kaggle ASL Alphabet. All inference runs on-device.

Currently exploring

LLM evaluation methodology, retrieval quality measurement, parameter-efficient fine-tuning (LoRA / QLoRA), and computer vision for accessibility.

Stack

PyTorch · HuggingFace Transformers · PEFT · sentence-transformers · ChromaDB · Ollama · FastAPI · Next.js / TypeScript · MediaPipe · ONNX Runtime · PostgreSQL · Docker

Microsoft Certified: Azure AI Engineer Associate · AI Fundamentals · Data Fundamentals

Reach me

Pinned Loading

  1. ASL-Translator ASL-Translator Public

    Real-time American Sign Language to text + speech translator that runs entirely in the browser. Phase 1: alphabet (in progress).

    TypeScript

  2. comic-scanner comic-scanner Public

    HTML

  3. Lyric-Recognizer Lyric-Recognizer Public

    Sing or speak song lyrics — get a ranked list of matching songs with Spotify and Apple Music links. Expo + React Native, web today, iOS next.

    TypeScript

  4. paper-tldr paper-tldr Public

    QLoRA fine-tune of Llama 3.2 3B Instruct for scientific abstract -> TL;DR summarization. Trained locally on RTX 4070 Ti. Eval-driven with prompt-sensitivity ablation.

    Python

  5. PaperPal PaperPal Public

    Chat with your research papers — fully local. RAG over PDFs with Ollama + Llama 3.1 8B, ChromaDB, FastAPI, Next.js 16. No API keys, zero cost.

    Python

  6. StreamScout StreamScout Public

    iOS app to search movies and see which US streaming services have them. Powered by TMDB.

    Swift