Universities need proof, not hype
The portfolio matters because it turns a one-off event into a visible artifact students can revisit, refine, and show.
University case study
How Innovation Time Lausanne turned agentic AI into 18 finance agents
A 90-minute campus workshop at UNIL where finance-curious Bachelor and Master students moved from AI curiosity to verified, tool-ready agent instructions they could keep using after the session.
Agentic AI, MCP and market finance, Innovation Time Lausanne x AI Workshop, Internef 122, UNIL, 13 May 2026
Case snapshot
From a student workshop to a public agent portfolio
The session was designed as a practical bridge between finance education and the new agentic workflow stack. The goal was not to impress students with AI demos. The goal was to leave behind a reusable professional artifact: agent roles, guardrails, authorised sources, output formats, and verification steps.
- Partner
- Innovation Time Lausanne
- Format
- 90-minute workshop
- Audience
- Bachelor + Master students
- Output
- 18 agent instructions
The UNIL student association for innovation and entrepreneurship.
A compact Agentic AI, MCP and market-finance session built around a live design sprint.
Pre-workshop survey responses showed 12 Bachelor and 3 Master participants.
A public GitHub portfolio of anonymised, deployable finance-agent specifications.
Audience evidence
The cohort was already using AI, but needed structure and verification
The pre-workshop questionnaire made the design challenge clear. These were not passive beginners: most already used ChatGPT or Claude. The gap was knowing how to turn everyday AI use into a reliable finance workflow with sources, limits, and human judgment.
12 / 3
Bachelor / Master
The education mix called for plain language, high pace, and professional examples without assuming technical depth.
9 UNIL / 5 EPFL
University signal
Institutional email domains showed a strong UNIL base with EPFL participation and one external address.
14 ChatGPT / 13 Claude
AI tool baseline
The room already had AI habits, which made the workshop focus on better workflows rather than basic chatbot discovery.
15 / 15
Finance-use-case demand
Every survey respondent selected market-finance use cases as a topic of interest.
Survey figures are aggregated from the pre-workshop questionnaire and exclude names and personal email addresses.
The brief: make agentic AI concrete for finance-curious students
The strongest student questions were practical: how AI can support market analysis, investment research, portfolio comparison, signal monitoring, and better financial decision-making without pretending to replace judgment.
The survey also exposed the trust problem. Several participants said their main doubts were hallucinations, judging answer quality, or not knowing how to use AI concretely. That is why the workshop put verification, authorised sources, and refusal rules at the center of the experience.
The design response: turn uneven confidence into a shared method
The session used AI Workshop's agent design method, the same framework used in corporate sessions across Switzerland, but compressed for a campus room:
- Feel: define the decision that must remain human
- Frame: choose a useful, bounded analysis case
- Imagine: formulate a single mission and the refusals
- Design: choose the sources, tools, and deployment surfaces
- Verify: define the output format, checks, and limits
Participants worked with AI Workshop's Agent Design Method Cards and structured worksheets, moving from a vague finance idea to a complete, deployable specification with authorised sources, explicit refusals, and a verification checklist.
In the room at UNIL
Facilitation, one-to-one mentoring, and open Q&A during the Agentic AI in Finance session
Materials and method in action
The Finance Agent walkthrough, individual design worksheets, and the Agent Design Method Cards in action
The output: 18 published, deployable agent specifications
By the end of the session, the group had collectively designed 18 AI agent specifications for market finance, each one anonymised, harmonised, and published as an open portfolio on GitHub. Among them:
- A financial ratio calculator that turns filings into readable ratios and an auditable summary
- A portfolio scenario comparator built around risk preferences and investment horizon
- An expert opinion comparator that preserves the nuance of conflicting analyst theses
- A market surveillance agent paired with a red-team counter-analysis to fight overconfidence
- A crypto watch agent that prepares buy/sell reflection without ever automating the decision
Every specification includes the agent's mission, authorised data sources, explicit refusals, an auditable output format, and ready-to-paste system instructions for ChatGPT Projects, Claude Projects, Claude Code, Codex, Antigravity, or MCP workflows. Participants received a certificate of attendance and a follow-up deployment guide that pushed one core habit: connect only trusted sources and verify outputs against the original source.
None of the agents produce buy or sell orders, guarantee performance, or make final decisions. Human judgment stays where it belongs: with the human. That constraint is not a footnote, it is the core of the method.
What made this session work
18 agents in 90 minutes
Every participant designed a complete agent specification. The full portfolio was published on GitHub within a day of the session.
Guardrails from minute one
Sources, refusals, verification checklists, and human decision points were designed into every agent, not bolted on afterwards.
Tool-agnostic by design
Each specification deploys in ChatGPT Projects, Claude Projects, Claude Code, Codex, or MCP workflows. Participants pick the tool they already use.
Real finance use cases
Ratio analysis, portfolio comparison, sector research, market monitoring. Cases participants chose themselves, not textbook examples.
A lasting public artifact
The open GitHub portfolio gives participants, and anyone else, reusable prompts and a professional trace of the session.
Human decisions stay human
No agent in the portfolio automates an investment decision. The method draws that line explicitly, every time.
About the facilitator: practical AI training across Switzerland
George Raymond-Alshoufi is the founder of AI Workshop Switzerland and one of the country's most hands-on AI trainers. His work focuses on a single gap: companies and institutions have powerful AI tools, but their people lack the habits, workflows, and confidence to use them in real work.
The method behind this session is the same one AI Workshop delivers to corporate teams across Switzerland, and the same one recognised by the Swiss Innovation Challenge 2026 jury for addressing a current and relevant problem in the practical introduction of AI in companies. From university classrooms in Lausanne to finance, HR, and operations teams in Zurich, Geneva, and Basel, the format adapts while the principle stays constant: participants build, verify, and leave with something they can deploy the next day.
AI Workshop Switzerland is a Microsoft Partner and a member of Anthropic's Claude Partner Network.
Reflection
What this case study proves
AI fluency is uneven
The room mixed confident AI users with students who had only heard of ChatGPT. The method cards gave everyone the same operating language.
Finance requires restraint
The strongest design choice was the refusal line: agents may analyse, compare, and brief, but the investment decision stays human.
Frequently asked questions
What is an agentic AI workshop?
A hands-on session where participants design AI agents: AI systems with a defined mission, authorised sources, guardrails, and human decision points. Instead of just prompting a chatbot, participants learn to specify agents that reason, verify sources, and work with real data, then deploy them in ChatGPT Projects, Claude Projects, Claude Code, or Codex.
How long does a session like this take?
This session ran 90 minutes. Corporate formats range from a 90-minute introduction to a full workshop day that includes a department assessment, an agent design sprint, and an AI roadmap.
Do participants need technical skills?
No. The method is built for non-developers. Participants work with structured worksheets and method cards, and the resulting specifications work in everyday tools. Advanced participants can deploy the same specifications in Claude Code, Codex, or MCP workflows.
Can our university, student association, or alumni network book a workshop?
Yes. AI Workshop Switzerland delivers the same agent design method on campus across Switzerland, adapted to your faculty, club, or alumni chapter: finance, entrepreneurship, HR, marketing, operations, or general business.
Book a workshop for your university or alumni network
The same hands-on method, adapted to your campus, your student club, or your alumni chapter. In person across Switzerland.