Practicum 2264 | Professor David Trevino

CGCS at ACC

Stefano Casafranca

Strategic Planner, Citizenship and Tech Initiatives

Product Designer AI Engineer

Background

Who I Am

Lima, Peru

Born and raised. Product design background.

Austin, Texas

Moved here to pivot into the tech scene. Graduating from ACC, May 14th.

What Drives Me

Products / AI systems that meet real-world institutions and serve real people at scale, reliably with the less friction and in the shortest time possible.

The Organization

Center for Government
and Civic Service

CGCS
At Austin Community College
16+
Partners: non-profits, City of Austin, LangChain, Austin Forum of Technology
Director
Bryan Port, my supervisor

CGCS trains and empowers students to be impactful civic leaders. Its initiatives provide students the skills, confidence, and sense of belonging to succeed in meaningful careers.

My Role

What I Actually Did

AI Agent Engineering

Claude Code, Codex, Gemini CLI, LangGraph orchestration, LangSmith observability

AI-Assisted Project Management

Obsidian as structured memory, wiki-as-router imitating vector RAG for decision-making

White Papers and Reports

AI environmental impact report for state leadership, weekly activity reports to VP

CGCS Event Intake Automation

In progress. CLI best practices and coding concepts hard to grasp for a non-technical team.

Program Design

16-week AI internship: NACE-compliant syllabus, recruiting pipeline, team structure

Web Development

CGCS website from zero in Astro: event pages, programs, public content

Events and Partnerships

AAIA Summer Intensive, SerpApi, Cloudflare, AI Alliance, Handshake pipeline

My Stack

The Three Tools That Made It Work

Claude Code
Anthropic Claude Code

AI agent engineering, autonomous coding, multi-file orchestration

Obsidian Master Board
Obsidian Obsidian Master Board

Structured memory, kanban tracking, wiki-as-router for decisions

LangSmith Observability
LangChain LangSmith Observability

Trace every agent run, debug failures, measure reliability

Flagship Project

GAPS Internship

Generative AI in Public Service

16
ACC students per cohort
16
Week structured program
4
Interdisciplinary teams

First Client: ACC Honors Program

User research to find pain points, then build an AI tool or agent that reliably solves them. Students build real tools for real people.

I wrote the NACE-compliant curriculum, the recruiting pipeline, and the syllabus. Launching Fall 2026.

The Good

What I Liked

Ownership, not observation

I co-led a program: NACE compliance, syllabus, recruiting pipeline, info sessions. My name is on the work.

Cross-functional range

Monday: white paper for state leadership. Wednesday: building the CGCS website. Friday: running a student info session.

Access to real decision-makers

Reported to a VP, sat in rooms with department heads, learned how institutional decisions actually get made.

Impact that outlasts me

GAPS launches Fall 2026 with real students and real clients. The Honors Program gets an AI tool built from user research. Tangible outcome.

The Real

Challenges and How They Got Solved

1

Institutional Pace + Coordination Tax

Work takes 50%. Waiting for next steps and connecting with the right people takes the other 50%.

Solved by: My Boss

Bryan had the authority to unblock bottlenecks: IT, cross-department approvals, finance.

2

Extreme Context Switching

Website, curriculum, events, partnerships, recruiting in the same week. No system to track it.

Solved by: Obsidian

Structured vault: dashboards, kanban boards, wiki-as-router. Every task and decision in one place.

3

Keeping Up with CLI Engineering

The role was mostly program design. Staying technical required effort outside work hours.

Solved by: Austin LangChain Meetups

Free, open to anyone, hosted at CGCS. Visit cgcs-acc.org for the next one.

Reflections

What Classes Did Not Teach Me

1
CLI agents changed how I see the future of work. Every company I build, I will automate with frameworks proven for human orgs, then translate them to agents.
2
AI in government is a blue ocean. Agencies waste time debating concerns. These tools can solve real problems instead.
3
Agents become reliable through labeling and rewards, not magic. Document 50/50 what went right and wrong. That feedback loop is how agents improve over time.
4
The goal is deterministic rules, not more AI. Use AI for the code to evolve to the point it does not need AI anymore. Just rules that make hard tasks repeatable.
5
Problems in any company repeat. We think every issue is new. Document them and you see the same problems repeat. That is why rules work.

Conclusion

Learn What I Learned

Apply to the GAPS Internship

Non-paid. 16 weeks. Build real AI tools for real public-sector clients. Fall 2026.

Info Session: May 15 at Highland

12:00 PM and 4:00 PM, in front of the Make It Center

Info Session: June 3, Online

12:00 PM and 4:00 PM

cgcs-acc.org/gaps
QR Code to GAPS application

cgcs-acc.org/gaps

Thank you. Questions?