All work04 / 04

Five projects · 2024 – 2026

Applied AI & Large Language Models

What does it take to put a language model inside a working product?

Schematic illustration · not project data

Scope
Course, client, and thesis projects
Period
Feb 2024 – May 2026
Methods
LoRA fine-tuning, retrieval-augmented generation, embeddings and retrieval, agentic workflows
Stack
Python, Hugging Face Transformers, OpenAI API, LangChain, ElevenLabs API

Projects

  1. SDSU, CS 668 · Jan – May 2026

    Fine-tuning Mistral 7B for product matching

    Fine-tuned Mistral 7B with LoRA to decide whether two product records describe the same item.

    0.33 → 0.91F1 on the product matching task, before and after fine-tuning

  2. SDSU, CS 668 · Jan – May 2026

    RAG assistant for the graduate bulletin

    Built a retrieval-augmented assistant that answers questions over the SDSU graduate bulletin.

  3. SDSU, CS 668 · Jan – May 2026

    Voice-based oral exam agent

    Built a voice agent on the ElevenLabs API that asks the exam questions and scores the answers.

  4. Lufthansa Industry Solutions · Sep 2025 – Jan 2026

    AI tender generation app

    A web app on a Python backend and the OpenAI API that drafts tender documents.

    Added a self-correction loop and automatic scoring that catch weak output before it reaches the user.

    ≈ 70%less manual tender drafting effort

  5. B.Sc. thesis with Otto Group one.O · Feb – Aug 2024

    Product matching framework

    A two-stage pipeline. Embedding-based retrieval with fine-tuned SentenceTransformers narrows the candidates, then a classifier decides the match.

    > 75%of products matched correctly