AI Chatbot Challenge:Innowing Trivia
Build a chatbot that can see beyond the screen.
Build your way to the frontier.
AI Chatbot Competition: Innowing Trivia is a competition and workshop programme where you learn to build increasingly capable chatbots, from simple language-model applications to systems that can retrieve information, understand images, reason across sources, and explore the physical world.
Registration closes
September 8, 2026.
Make Innowing easier to discover.
Your chatbot should help students find information and opportunities at the Innovation Wing.
- Programmes, activities, and opportunities at the Innovation Wing
- Information on the Innowing website, including equipment, membership, etc.
- Information collected from designated areas in the building
Five evaluation levels
| Level | What it tests | Example question |
|---|---|---|
| Lv 1 | General Knowledge Simple queries solvable by the LLM's internal knowledge base. | What faculty does the Tam Wing Fan Innovation Wing belong to? Engineering |
| Lv 2 | Text Retrieval Specific facts buried in website text data. | When was the deadline for the Funding Scheme 1st round submission in 2025-26? October 17, 2025 |
| Lv 3 | Visual Retrieval Information found only in images on the websites. | What were the winning teams in Pitch New Tech Ideas in 2025? Smartsocks, Digitio, Gingtrolley |
| Lv 4 | Reasoning Requires reasoning from data, or synthesizing information across pages. | How many workshop posters are on the Innovation Academy websites for 2025? 31 |
| Lv 5 | Physical World RAG Questions about the physical layout or signage in Inno Wing. | On the wall at the back of the brainstorming area, there are three words. HKU, Engineering, and...? Innovation |
What you will practise
Work with language models
Start with prompt engineering, model calls, and clear response formatting.
Retrieve text and visual evidence
Prepare website content and images so your chatbot can find the information it needs before answering.
Reason across clues
Combine information from different sources and work with knowledge collected from the physical Innovation Wing.
Why participate?
Turn introductory AI knowledge into a working chatbot. Test your ideas through workshops, experiments, and evaluation.
Work on a real Innovation Wing problem, learn from optional peer mentors, and potentially be considered for advanced GenAI or R&D projects.
Starter materials
Start with the repository and build your own full pipeline.
Starter repo
The competition repository with baseline implementation, workshop materials, data files, and testing utilities.
Open starter repo ↗Workshops and tutorials
Computer Setup Guide
- Self-paced on Notion.
- Covers setup, repository access, credentials/configuration, API access, and acceptance test.
Important reminder: The acceptance test verifies that your setup is successful. You must pass this test before Workshop 1. If you have difficulties with computer setup, you are advised to join the Computer Setup Help Desk session.
Computer Setup Help Desk
- Optional drop-in support before Workshop 1.
- Staff will provide assistance onsite at Makerspace A, Innovation Wing One.
- Participants shall bring their own computers to join this session.
- Get help with software environment, repository setup, credentials, API access, and setup acceptance test.
Workshop 1: RAG Pipeline
Understand RAG and hallucination, use embeddings and vector databases, build an index, experiment with retrieval, and evaluate the result.
- Live at Makerspace A, Innovation Wing One
- Participants shall bring their own computers to join this session.
- Covers Lv1 and Lv2
- Optional 30-minute mentor clinic afterwards
Code Walkthrough Online Tutorial
- Recorded and self-paced on Notion.
- Covers the code on the starter repo, including the
main.pycontract and baseline implementation of a workable RAG chatbot.
Workshop 2: Visual and Physical Data
Describe and index image content, compare visual retrieval approaches, decompose complex questions, use metadata filters and aggregation, and prepare for physical-world data collection.
- Live at Makerspace A, Innovation Wing One
- Participants shall bring their own computers to join this session.
- Covers Lv3, Lv4, and Lv5
- Optional 30-minute mentor clinic afterwards
Checkpoints
Small milestones help teams test ideas while there is still time to improve.
Get the team, tools, credentials, and setup ready.
Sep 1–16, 2026
Registration closes.
Inno Wing API keys and Computer Setup Guide are released.
Checkpoint 0 deadline: complete computer setup.
Optional computer setup help desk.
Build your RAG chatbot.
Sep 16–23, 2026
Workshop 1: RAG Pipeline.
Optional Code Walkthrough recording released after Workshop 1.
Checkpoint 1: own scraper and vector database; score above zero on Lv1 and Lv2.
Add physical-world intelligence capability.
Sep 30–Oct 14, 2026
Checkpoint 2: diagnostic experimentation with image data.
Workshop 2: Visual and Physical Data.
Checkpoint 3: collection log entry for assigned zones.
Checkpoint 4: image descriptions indexed; score above zero on Lv3.
Competition Day!
Oct 21–28, 2026
Deadline for preliminary round submission (online).
Final round competition (Makerspace A, Innovation Wing One).
Rules and Eligibility
Who can join
- Core team of 2–3 undergraduates in Years 1–3.
- Optional: one Year 4-or-above undergraduate peer mentor.
- Basic Python knowledge equivalent to COMP1117/ENGG1330.
- Up to 20 teams accepted; 5 advance to the final.
Register by Sep 8, 2026
Form your team and complete the registration process before the deadline. Mentor nomination details: TBC
Open registration ↗Permitted knowledge sources
Teams may use:
- Tam Wing Fan Innovation Wing (One) websites ↗
- Innovation Academy websites ↗
- Information found in designated Innovation Wing areas:
- Photo gallery near the entrance
- Makerspace A
- Brainstorming area
- Open event area
- Digital Learning Studio
The competition knowledge is static during the competition. Teams must collect and prepare their own corpus in advance.
Allowed
AI models
- GPT-5-mini
- GPT-4o-mini
- DeepSeek-R1
- Qwen-3 Plus
Embedding models
text-embedding-3-small- Permitted local embedding models
Not allowed
- Use external APIs during the competition
- Live web searches
- Modifying the code or database during runtime
Your chatbot logic goes here.
# team chatbot logic goes here
return answer
The chatbot will be run by executing rag_answer() in main.py. Do not change other functions in that file.
Detailed submission instructions
The final submission process, required files, execution procedure, response format, timing limit, dependency installation, and final checks will be published here when confirmed.
Judging and Prizes
Prizes and Benefits
20 questions
Teams submit to our system by Oct 21, 2026. An automatic script runs the questions against each chatbot, and judges grade the answers manually.
Lv1 General knowledge · 30%
Lv2 Text retrieval · 30%
Lv3 Visual retrieval · 40%
20 questions
Best five teams in preliminary round qualify. Questions are asked and graded on the spot, with substantial Lv4 reasoning and Lv5 physical-world RAG content.
Lv1–Lv3 combined · 30%
Lv4 Reasoning · 30%
Lv5 Physical-world RAG · 40%
1
Correct and complete
Factually correct and complete answer.
0.5
Partially correct
Awarded at judges’ discretion.
0
Non-responsive
Incorrect, irrelevant, or non-responsive.
Sudden Death → Preliminary Ranking
Additional 5 questions from any level will be used to resolve a tie in a sudden-death format. If all five questions are used and the tie remains unresolved, the team that ranked higher in the preliminary round will be placed higher.
