Skip to content

CSE120: The Software Engineering Capstone

CSE120 is the senior capstone course in Computer Science & Engineering at UC Merced. With 114+ students per semester and teams building real software systems for external stakeholders, it serves as the CSED Lab's primary research vehicle for CS pedagogy at scale.

The lab doesn't just run the capstone — we study everything about it. Every lab assignment, every grading run, every team formation cycle generates data that feeds back into our research. The course is a living laboratory for understanding how students learn software engineering.

What We've Learned

AI-Assisted Grading

Traditional autograders work for multiple-choice questions, but capstone projects are open-ended by nature. Teams build different things in different ways. We developed a stage-decomposed LLM evaluation pipeline — Extractor, Validator, Scorer — that can evaluate architecture diagrams, requirements documents, and system designs alongside code. The pipeline processes student submissions through five phases: data ingestion, staged AI evaluation, post-hoc calibration, human-in-the-loop review, and Canvas export.

What we found: AI evaluation at capstone scale works best when treated as an assistant, not an oracle. The pipeline surfaces candidate penalties and evidence traces; human instructors make the final call. Post-hoc calibration detects when the AI is being too harsh or too lenient and adjusts accordingly.

Team Formation

Forming 20–30 balanced teams from 114+ students is an NP-Hard problem. We built a neuro-symbolic pipeline — SkillNet for skill extraction (GNNs over a computer science ontology) and OptiTeam for constraint optimization (Integer Linear Programming) — that replaces the manual "get into groups of four" approach with algorithmic team formation grounded in survey data and skill inference.

Course Design

The course has evolved from traditional "lecture + project" toward an Actionable Agenda model: 170-minute lab sessions structured around stand-ups, individual demos, and workshop time. Teams work on real industry projects with external stakeholders, follow PR-only workflows, and submit changelogs with semantic versioning rather than sprawling prose reports.

The course culminates in the Innovate to Grow (I2G) showcase event — a public presentation of all capstone projects to judges, industry partners, and the campus community.

For students enrolled in the course, the CSE120 Knowledge Base provides the day-to-day guide to assignments, resources, and workflows.

Explore

  • GradeFlow — Visual grading pipeline builder informed by CSE120's autograding architecture.
  • Team Evaluation — Canvas LTI peer evaluation platform used in CSE120.
  • T4 + TA Copilot — GitHub-integrated task tracking and TA assistant for capstone teams.
  • Capstone Survey — Collects student project preferences and skill profiles each semester.
  • Psephos — Survey and polling API underlying the capstone survey platform.
  • UniPoll — Classroom polling web app built on Psephos.
  • UMACS — Centralized user management across the CSE120 tool suite.