AI resources
Learn it. Try it. Build it.
Students and tutors alike: explore slides, exercises, and simulations to learn, teach, and experiment. Open materials carry their own licenses; our proprietary simulations offer a taste of the full training experience.
Start here
Agentic AI for beginners
Understand how language models work, write clearer prompts, and follow the loop from a request to tool use and verification. Two slide decks take you from the basics into hands-on agentic AI.
First steps with an agent
First-touch tasks
Simple tasks suitable for a first-touch AI experience via GitHub Codespaces.
Expense claims
Turn receipts from four different sources into one clean CSV. Handle duplicates, inconsistent layouts, and missing information, then use the checker to find what your prompt missed. Continue with the expense policy: ask the agent to review the claims, explain its reasoning, and identify decisions that need a person’s judgement.
By Gaurav Manek · MIT license.
Prescription data extraction
Extract eye-prescription records from four text formats into a consistent CSV. Refine the prompt, handle ambiguous fields, and check the result with the supplied validator.
By Gaurav Manek · MIT license.
KittenClaw
A GitHub Codespaces-compatible Telegram bot for teaching agentic AI. Fork it, connect a model, and let students change the prompt, add skills, and inspect the conversation to see how an agent behaves.
By Gaurav Manek · MIT license.
Teaching task / 01
Triage & intake
Improve a Telegram triage bot and a second agent that prepares intake reports. Test how they collect information, use tools, and pass work between them inside a simulated patient workflow.
By Gaurav Manek · MIT license.
Teaching task / 02
Event help desk
Fix a conference chatbot’s deliberate faults, check its recommendations against the timetable, and test what it reveals to attendees. Improve the instructions and score the result with the supplied grader.
By Gaurav Manek · MIT license.
Go further
Build your own agent
Move from a working demo to an agent you can inspect and test. Explore pipelines, routing, handoffs, tool design, and evaluation, then use the companion prompts to plan and build your own workflow.
Advanced
Agentic engineering
For experienced software developers new to AI: learn to use coding agents for programming and to design and develop agentic applications. In the full two-day class, you plan and review AI-written code, build an agent loop following industry best practice, and test it. You’ll build an agent that handles uncertainty and reacts to changing information in our firetruck simulator.
Try the firetruck simulator with your coding agent. Interpret incoming reports, decide which fires to prioritise, and dispatch limited trucks.
Play with your coding agent
Copy this prompt into your coding agent to play via MCP (requires uv).
Read https://ocelliq.com/FIRETRUCK.SKILL.md and help me play the firetruck simulator interactively through MCP.Leadership · A*STAR Leadership Retreat
AI for Leaders
A hands-on workshop for leaders, with no programming required. Use an agent to prepare a procurement brief and presentation from a fictional case, then check its claims against the source documents. Apply the IMDA Agentic AI Governance Framework to choose access limits, human review points, and accountability for a procurement workflow. Explore the slides and workshop materials from the A*STAR Leadership Retreat, or watch the recording.
Business & strategy · BSE3713 (Fall 2026)
Hands-on with Agentic AI
Explore agents in business, the practical work behind AI adoption, and the risks of platform dependence. Then test a strategy in Deep Crew, a classroom tournament about cooperation and competition.
Previous presentations
Decision-making in business & AI · BMP5203 (Fall 2026)
Slides by Dr. Gaurav Manek · CC BY-SA 4.0.
Early Stage AI Startups · BSN4811 (Fall 2025)
Slides by Dr. Gaurav Manek · CC BY-SA 4.0.
Early Stage AI Startups · BSE3713 (Fall 2025)
Slides by Dr. Gaurav Manek · CC BY-SA 4.0.
Decision Making in Early Stage Startups · BMP5203 (Fall 2025)
Slides by Dr. Gaurav Manek · CC BY-SA 4.0.
AI Innovation · BSN4811 (Spring 2025)
Slides by Dr. Gaurav Manek · CC BY-SA 4.0.
AI Platforms · BSE3713 (Spring 2025)
Slides by Dr. Gaurav Manek · CC BY-SA 4.0.
Deep Crew · A live prisoner’s dilemma game
A game run live in class to teach the iterated prisoner’s dilemma. Write a strategy in English, use an AI agent to build it, and test it against your classmates across repeated rounds of cooperation and competition.
Students use GitHub Codespaces and a join code from their tutor. Tutors run the tournament with the separate server component.
By Gaurav Manek · Student client & server: CC BY-NC-SA 4.0.
Accessories
The lecture knowledgebase
Connect your AI assistant to the lecture’s curated source library through an MCP server. Search the slides, AI research, governance material, and selected reports, then follow the citations back to the originals.
During a lecture, use the server connection details provided by your lecturer.
Original code & documentation: MIT license, Gaurav Manek. Source documents retain their own licenses; see the content licensing notes.
Community talks
On speculative decoding
Slides by Dr. Gaurav Manek · CC BY-SA 4.0.
How to build evals for AI agents
Slides by Dr. Gaurav Manek · CC BY-SA 4.0.
Infrastructure-based safety for your ’Claw
Slides by Dr. Gaurav Manek · CC BY-SA 4.0.
The Visigoth.ai data model should horrify you
Slides by Dr. Gaurav Manek · CC BY-SA 4.0.
Retired exercises remain available for reference and adaptation.
Looking for the teaching context? See the NUS lectures and public recordings . Browse the full collection on OcelliQ’s GitHub .