Dacheng Shen | Academic Homepage

🎓 Education

University of Southern California June 2024 - Dec 2025

M.S. in Computer Science

University of Connecticut Sept 2020 - May 2024

B.S. in Computer Science

Dean's List Honors

Software Development Engineer Jul 2023 - Aug 2023

HICOCA Intelligent Equipment Technology

Comparative Narrative Analysis with Claude 3.7 (Research Assistant) Mar. 2025 - Now

Designed and automated a full Claude 3.7 Sonnet-based LLM evaluation pipeline for 208 narrative pairs under four comparison dimensions (conflict, unique, holistic, overlapping)
Engineered scalable prompt-based inference using anthropic API and batched output parsing, reducing manual judgment workload by over 90%
Structured model responses into JSON and human-readable formats to enable systematic cross-model analysis and future human annotation
Collaborated with faculty and research team on dataset preparation, narrative extraction, and designing metrics for minimizing bias in LLM comparison

OEM Controls Automated Angle Test for AS5 Aug. 2023 - May. 2024

Led a year-long capstone project to automate Quality Assurance testing for the AS5 sensor system, significantly improving testing efficiency and reliability
Developed a Python-based control system on Ubuntu to operate a 6-axis robot arm, automating sensor calibration and test procedures with precise "Pass/Fail" feedback and real-time result logging
Designed and implemented a seamless UI-Backend integration using APIs, replacing the previous client version with a fully automated calibration solution
Successfully delivered the project, reducing manual testing time by 67% and improving calibration accuracy by 26%, which enhanced overall efficiency and reliability

Multi-Backbone Transfer Learning for Fine-Grained Solid Waste Classification Jan. 2025 - May 2025

Developed a multi-class image classifier for nine waste categories using Python, TensorFlow/Keras, with an 80/20 train/test split and 20% in-training validation for robust model evaluation
Built a preprocessing pipeline to resize images and applied extensive data augmentation (random crops, zooms, rotations, flips, contrast adjustments, translations) to mitigate overfitting
Implemented transfer learning by freezing pretrained backbones (ResNet50, ResNet101, EfficientNetB0, VGG16) and adding custom FC layers with ReLU, softmax, L2 regularization, batch normalization, and 20% dropout; trained with Adam optimizer and early stopping over 100 epochs
Evaluated all models on test set using precision, recall, AUC and F1-score; identified VGG16 as the top performer—achieving highest accuracy and generalization—and selected it as the final model

AI-Driven Travel Assistant Design – FlySmart Mar. 2025 - May. 2025

Led a requirements engineering project to design an AI-powered flight booking assistant
Conducted stakeholder analysis, surveys, and interviews, derived user personas, empathy maps, and categorized requirements
Created and validated a Figma prototype with features including flight search, visa management, smart alerts, and personalized AI chat interface
Applied agile methodology with story-driven backlog, sprint-based development roadmap, and formal validation via usability testing and prototyping

Semantic Retrieval and QA System with Weaviate & RAG Sep. 2024 - Dec 2024

Built a semantic search pipeline using Weaviate with text2vec-transformers for vector-based brand similarity via GraphQL nearText queries
Developed a PDF QA chatbot with Streamlit, leveraging PyPDFLoader, ChromaDB, and Hugging Face embeddings in a RAG framework for document-grounded response generation

Student Admin Design Sep. 2023 - Dec. 2024

Collaborated on a group project, using Figma to design the product model and present the plan to stakeholders
Developed and implemented the code to bring each Figma design feature to life
Completed the website development and thoroughly tested all functionalities

Food Ordering App Design Aug. 2021 - Nov. 2021

Used Java to build a food ordering app that could generate an invoice with the name and price of the dish and the total price after ordering
Group leader, used GitHub to share and merge information and parts of what each group member had finished
Completed a report and made a presentation

Languages

Python

Java

C

C++

SQL

HTML

Tools

Git

AWS

Docker

Figma

Trello

Technologies

RAG

Hugging Face

Web Crawling

OOP

Relational DB