A one-day GenAI workshop syllabus for colleges (hands-on, no fluff)

A one-day GenAI workshop designed for colleges focuses on hands-on learning, ensuring students leave with a functional project and a method to evaluate its performance. The plan is structured for 30 to 60 students, each equipped with a laptop and a basic understanding of Python and a code editor. Prerequisites are sent out a week in advance, including installation of Python and setup of a code editor, along with a ten-minute self-check to ensure everything is working correctly.

Participants need a free account on an LLM provider or will have keys supplied by the organizers for the duration of the workshop. Each student receives a PDF containing vital notes, manuals, and a rulebook. The workshop begins with an introduction to language models, demonstrating their use through live calls from Python. Students make three calls to a model, with one call being broken deliberately. There are no slides during the first hour, focusing solely on practical application.

The second block, also 90 minutes, delves into retrieval-augmented generation. Students are guided to chunk their own PDF documents, embed the chunks, and use them to search and generate responses within the prompt. The exercise vividly illustrates the impact of context on model output, as seen when hallucinations are reduced with proper context and reappear when information is missing. This block sets the stage for the afternoon discussions on enhancing model honesty.

In the third block, lasting 60 minutes, students learn to make their models more reliable. They incorporate keyword search alongside vector search, set thresholds to reject weak matches, and add citations referencing the exact page of retrieved information. This round equips students with strategies to improve confidence in model outputs within an hour.

The fourth block, also 60 minutes, focuses on evaluation and deployment. Each student writes ten questions with known answers, scores their system's performance, and deploys a minimal version of their model, securing a URL for demonstration. The workshop concludes with students sharing their links and scores, emphasizing performance metrics over formal certificates.

Post-workshop, participants receive the complete code and a written version of each block's content. An optional stretch task, completed within a two-week period, involves students adding reranking or swapping the embedding model, reporting the impact of these changes. A remote office hour is scheduled two weeks after the workshop to address any remaining questions.

This syllabus was crafted by Pranjul Rathour, a GenAI engineer from Kanpur, India, currently serving as the CTO at SCULT INDIA. Rathour is responsible for deploying production RAG, fine-tuning, and agentic AI systems, mentoring over 200 students through TechVerse Enclave, and judging/ speaking at student hackathons nationwide. For inquiries about GenAI workshops, mentorship, or speaking opportunities, contact Pranjul at pranjulrathour41@gmail.com.

Additional resources, including his portfolio, blog, and social media profiles, are available on his website and linked platforms.