Operable
Entries must be systems that actually run. The final is held as an on-site working demo with presentation, Q&A, and showcase.
MOE ADVISED・NAPAI HOSTED・WITH TAAI 2026
OPERABLE・MEASURABLE・CONTROLLABLE・FIELD-VALIDATED
NAPAI Agentic AI Innovation
Competition for Campus Applications
Solving real campus and
field problems with Agentic AI.
University students nationwide are invited to start from user needs and build intelligent applications that reason, plan, call tools, integrate systems, and adjust their actions based on execution results. Entries must be operable, measurable, controllable, and validated in the field.
COUNTDOWN TO THE FINAL
Final & Showcase・2026.11.21・Asia/Taipei
00
Days
00
Hrs
00
Min
00
Sec
01 / ABOUT
About the competition
The competition centers on real fields setting the problems, student teams solving them with Agentic AI, and results that are operable, measurable, controllable, and validated for deployment. Students nationwide are invited to build innovative work with technical depth, practical value, and potential to spread.
Once AI enters campus, industry, and public service settings, it still faces need discovery, process integration, data governance, permission control, system reliability, and dynamic environments. The Ministry of Education's National Agentic & Physical AI Initiative (NAPAI) focuses on bridging the gap between models, simulation, and the real world. Advised by the MOE Department of Information and Technology Education, coordinated by the Department of Computer Science and Information Engineering at National Central University and the Graduate Institute of Information Management at National Taipei University, and held with the 31st International Conference on Technologies and Applications of Artificial Intelligence (TAAI 2026).
Measurable
Teams must provide test cases, baselines, and metrics, and disclose success rates, latency, cost, error types, and execution traces.
Controllable
High-risk actions require human confirmation, tiered permissions, and complete logs, with prompt injection, data leakage, and unsafe output assessed.
Field-validated
Evidence of need and user feedback follow the field validation levels, alongside deployment, adoption, operations, cost, and diffusion plans.
02 / ELIGIBLE WORK
DEFINITION AND MINIMUM REQUIREMENTS
What qualifies as an entry
Agentic AI here means a goal-oriented AI system that plans, decides, calls tools, or interacts with systems according to task state, and adjusts subsequent actions based on intermediate results, errors, or user feedback. An entry must meet the first item below, plus at least two of the remaining three.
Goal orientation and task decomposition
- Two or more steps
- Steps with sequential dependencies
- Progress driven by execution state
Tool or external system invocation
- APIs and databases
- Knowledge bases and campus systems
- Software tools, cloud services, or devices
State management and dynamic adjustment
- Revise plans on intermediate results
- Retry or switch to alternative tools
- Abort the task when needed
Permission and risk control
- Human confirmation and permission limits
- Operation logs and failure recovery
- Output checking
Single-turn Q&A, one-shot content generation, pure document retrieval Q&A, fully fixed automation without state-based judgement, or video simulation with no verifiable system will in principle not meet the minimum requirements. Multi-agent design, long-term memory, self-reflection, and self-repair are not mandatory; they should be adopted as the problem requires, and judges will weigh their appropriateness rather than the size of the technology stack.
03 / GROUPS & THEMES
EACH TEAM PICKS ONE GROUP AND ONE THEME
Competition groups
Campus Self-Proposed Group
Teams raise their own real problem in university teaching, learning, research, administration, services, or campus life, and secure their own interviews, process data, or test field.
Designated Field Group
Teams build against topics, data, APIs, or test environments announced by the organizer, partner schools, companies, or institutions. Each topic's limits and resources follow the competition website.
DESIGNATED CHALLENGE PARTNER
Ai3 Artificial Intelligence Co., Ltd. ↗
DESIGNATED APPLICATION FIELDS
Taipei City, New Taipei City, and Taoyuan CityDESIGNATED CHALLENGE
GovAid Copilot: Autonomous AI Agent for Childcare & Housing Subsidies
申請政府補助神隊友：育兒／租屋補助導航與合規申辦 AI Agent
CHALLENGE DESCRIPTION
Childcare and housing subsidies involve household-registration rules, household-income thresholds, dependent household members, eligibility requirements, and additional conditions. They provide a suitable test of an Agent's ability in multi-turn dialogue guidance, information collection, eligibility assessment, rule-based reasoning, and compliant application support.
Teams may design a voice or text Agent to help users clarify their circumstances, identify subsidy programs they may qualify for, and navigate the application process and required documents.
DATA SOURCES & VALIDATION SCOPE
Official websites of the Taipei City, New Taipei City, and Taoyuan City governments and relevant agencies—such as social affairs departments and housing authorities—are the primary sources for data collection, retrieval, and validation.
Entries should be based on the latest official subsidy rules valid in 2026. The policy versions, designated websites, data scope, and gold answers used for actual evaluation will follow later announcements on the competition website.
THE ORGANIZER MAY REASSIGN A TRACK BASED ON THE WORK ITSELF
Field themes
Smart Academic Affairs & Learning Support
Course selection, learning diagnostics, academic advising, teaching support, personalized learning paths, and learning outcome analysis.
Campus Administration & Service Processes
Forms, documents, policy search, cross-office applications, campus life services, and process automation.
Library, Research & Knowledge Services
Catalog and knowledge base retrieval, research exploration, literature synthesis, resource recommendation, and research administration support.
Industry & Public Field Co-Creation
Knowledge management, customer service, query analytics, process automation, IoT, or other applications proposed by partner companies, institutes, government, or non-profits.
Technical tags・multi-select
Agentic Workflow
Domain-Specific Agent
Multi-Agent Collaboration
RAG／Knowledge Agent
Physical／Digital Embodiment
Human-in-the-loop
Other tags
Technical tags are used for judging groups and outcome analysis only; they do not affect group or theme assignment.
04 / JUDGING
TWO STAGES・FROM DOCUMENT REVIEW TO A LIVE WORKING DEMO
Criteria and weightings
Preliminary
Document review of deck, video, and field materials
- Deck
- PDF, max 12 pages
- Video
- Max 4 minutes
- Deadline
- 2026.10.16
- Total
- 100 points
-
Problem and field evidence20%
Authenticity of the problem, understanding of users and processes, and quality of need validation.
-
Appropriateness of Agentic AI design20%
Task decomposition, decisions, tool invocation, state management, and why an agent is necessary.
-
Technical and data planning20%
Architecture, integration, data quality, permissions, deployment, and engineering feasibility.
-
Prototype and validation plan15%
Test cases, baselines, metrics, failure scenarios, and feasibility of completing the final prototype.
-
Innovation and diffusion potential15%
Novelty, degree of improvement to the field, and reuse across units, campuses, or industries.
-
Safety, governance, and ethics10%
Personal data, security, hallucination, bias, human review, traceability, and risk control.
Six criteria・total100%
Final
On-site working demo, presentation, Q&A, and showcase
- Finalists
- 2026.11.06
- Materials due
- 2026.11.13
- Final date
- 2026.11.21
- Total
- 100 points
-
Task completion and stability25%
Completion rate, accuracy, reproducibility, latency, and exception handling on assigned and self-chosen tasks.
-
Field impact and user validation20%
Improvement to the need, user feedback, before-and-after comparison, adoption potential, and sustained value.
-
Agentic depth and appropriateness20%
Autonomous planning, tool selection, state management, system integration, and sound technical choices.
-
Safety, controllability, and governance15%
Permissions, human confirmation, output checking, personal data and security, traceability, and failure recovery.
-
Innovation and diffusion value10%
Novel perspective, breakthrough for the field, and reuse potential across fields.
-
On-site demo and Q&A10%
Clarity of explanation, completeness of the operation, team command of the work, and quality of answers.
Six criteria・total100%
More autonomy is not automatically better. Where administrative approval, grades, finance, medical, legal, personal data, or other high-risk matters are involved, appropriate human review, tiered permissions, and abort mechanisms count as good design. Finalists must present verifiable results; teams that fail to attend, complete required documents, or present any verifiable result for reasons attributable to them may lose finalist or award status. Where force majeure, field limits, or third-party outages are involved, teams may submit evidence and, once approved, substitute recordings, offline environments, test records, or other means.
05 / SUBMISSIONS
PRELIMINARY MATERIALS・PRE-FINAL MATERIALS
What to submit
Teams register and submit through the designated system on the competition website before each deadline. Late, incomplete, corrupted, or inaccessible submissions may be rejected; correctable administrative omissions may be notified for a single round of correction. No document or code may contain passwords, API keys, non-de-identified personal data, or other unauthorized confidential information.
Required preliminary materials
- Proposal deck, PDF, max 12 pages
- Filename: group_theme_team_proposal.pdf
- Concept／demo video, max 4 minutes
- Field need and validation materials
- Online registration form and declaration
Pre-final materials for finalists
- Working prototype or test environment
- Code repository or encrypted source package
- README／technical documentation
- Agent evaluation data and one trace
- Updated deck and demo requirements
The proposal deck should cover the real problem and evidence of need, why Agentic AI is necessary, task flow and system architecture, state management and safety design, prototype and evaluation metrics, deployment and diffusion plans, and team roles and third-party resources. Preliminary review is in principle anonymous: the deck, video, and filenames must not reveal the school, department, adviser, or any identifying mark.
06 / FIELD VALIDATION
Field validation levels
The preliminary round expects at least Level 1. To score well on "field impact and user validation" in the final, teams should in principle reach Level 2 or above. Where sensitive data, security, or legal limits apply, de-identified data, synthetic data, a sandbox, or another organizer-approved alternative may be used.
Problem hypothesis
The team states a problem, with no target users or process evidence yet.
Need interviews
Target users, process owners, or stakeholders have been interviewed, with pain points and needs documented.
Data／process validation
De-identified data, process documents, test cases, API specifications, or other verifiable material obtained.
Prototype trial
Real users operate the prototype in a simulated, sandboxed, or controlled environment and give feedback.
Field pilot
Limited deployment in a real unit, with operation records and quantified before-and-after comparison.
Field data, accounts, APIs, devices, and documents may be used only for competition purposes within the authorized scope; they may not be published, transferred, reverse engineered, accessed without authorization, or used to train other models, and must be deleted or returned at the provider's request after the competition. Field providers may not use the competition to obtain a production or commercial system for free; later adoption, technology transfer, internships, or commercial cooperation require a separate agreement.
07 / ELIGIBILITY
Eligibility and team rules
Open to currently enrolled students at colleges and universities nationwide, including master's and doctoral students, in any major; enrollment status is determined as of the registration deadline. Teams warrant that their registration information is accurate and their work original, and agree to follow the rules, field requirements, and later announcements.
2–5
Two to five students per teamOne leader is required and one faculty adviser is optional; teams doing real field validation must also name a field contact, who may be the same person as the adviser.
＋
Cross-campus and cross-disciplineEach student may join only one team, and the same core work may not be entered by different teams or in different groups.
✓
Non-students may only adviseTeachers, field staff, and technical advisers may guide, explain needs, and help with testing, but may not carry out the core design, development, evaluation, or presentation.
↺
Changes require approvalChanges to members, leader, adviser, or field contact must be requested within seven days of the finalist announcement; members may only be removed or replaced, never added beyond five.
08 / AWARDS
TOTAL NT$150,000・NT$75,000 PER GROUP
Awards and what follows
The total prize pool is NT$150,000, split between the Campus Self-Proposed Group and the Designated Field Group at NT$75,000 each, with eight winning teams per group. All prizes are awarded to the team as a unit.
First place｜NT$25,000
One team in each group, NT$25,000 per team.
Second place｜NT$15,000
One team in each group, NT$15,000 per team.
Third place｜NT$10,000
One team in each group, NT$10,000 per team.
Honorable mention｜NT$5,000
Five teams in each group at NT$5,000 per team, NT$25,000 per group.
Any award may be left unawarded based on quality, and the judging committee may adjust the number of awards or move them between groups within the total prize pool. Prizes are subject to income tax withholding, and teams must provide receipts and required documents by the organizer's deadline. The organizer may add corporate, field, safety, cross-domain, or other special awards depending on sponsorship, with stacking rules announced per award. Finalists and winners receive a certificate of participation or of award.
09 / TIMELINE
Key dates
From the rules announcement, online registration, and the briefing and enablement workshop, to preliminary document review, finalist mentoring and prototype refinement, and finally on-site validation with a working demo at TAAI 2026.
Registration opens
FridayRegistration closes
Friday 23:59Briefing & workshop
Friday 14:00–16:00Preliminary deadline
Friday 23:59Finalists announced
FridayFinal materials due
Friday 23:59Final & showcase
SaturdayThe rules, registration form, and field topics are published on the competition website; the curriculum, materials, and format of the briefing and enablement workshop are announced separately; once the finalists are announced, those teams enter mentoring and prototype refinement. The organizer may adjust the schedule, format, or venue due to venue, natural disaster, epidemic, system, or other necessary circumstances; any change follows the announcement on the competition website.
10 / FINAL VENUE
NOVEMBER 21, 2026・NEW TAIPEI CITY
The Great Roots
Forestry Spa Resort
The final and showcase are held at The Great Roots Forestry Spa Resort, with the exhibition and awards alongside TAAI 2026. Presentation, operation, and Q&A times and equipment specifications will be announced separately; judges may ask teams to run designated test cases and may inspect system logs, documents, and code.
- Date
- Saturday, November 21, 2026
- Venue
- The Great Roots Forestry Spa Resort, Sanxia, New Taipei City
11 / FAQ
FAQ
Who can join and how large is a team?＋
Currently enrolled students at colleges and universities nationwide, including master's and doctoral students, in any major; enrollment is determined as of the registration deadline. Teams have two to five students with one leader and an optional faculty adviser, may be cross-campus and cross-discipline, and each student may join only one team.
What qualifies as an entry?＋
An entry must show goal orientation and task decomposition, plus at least two of tool or external system invocation, state management and dynamic adjustment, and permission and risk control. Single-turn Q&A, one-shot generation, pure retrieval Q&A, fully fixed automation without state-based judgement, or video simulation with no verifiable system will in principle not qualify.
What is required for the preliminary round?＋
A proposal deck in PDF of no more than 12 pages (cover, contents, and appendix all counted), a concept or demo video of no more than 4 minutes, field need and validation materials, and a completed online registration form with declarations. Review is in principle anonymous, so the deck, video, and filenames must not identify the team.
Is a real field partner required?＋
Validation is measured on levels 0 to 4. The preliminary round expects at least Level 1 (need interviews), and scoring well on "field impact and user validation" in the final in principle requires Level 2 or above. Where sensitive data, security, or legal limits apply, de-identified data, synthetic data, a sandbox, or another approved alternative may be used.
Is a working demo mandatory in the final?＋
The final is held as an on-site working demo with presentation, Q&A, and showcase, and finalists must present verifiable results. Teams that fail to attend, complete required documents, or present any verifiable result for reasons attributable to them may lose finalist or award status. Where force majeure, field limits, or third-party outages are involved, teams may submit evidence and, once approved, substitute recordings, offline environments, or test records.
Who owns the intellectual property?＋
Intellectual property in an entry belongs in principle to the team, unless a contract provides otherwise, which must be disclosed at registration. Teams grant the organizer free use of the work's name, abstract, images, deck, demo video, and publicly presented content as needed for promotion, showcase, education, and non-profit outreach; code, data, and confidential content may not be published without the rights holder's separate consent.
REGISTRATION ・ 2026.08.28
Registration
The rules, registration form, and field topics are scheduled for publication on the competition website on August 28, 2026, with online registration closing on September 11 at 23:59. Start forming your team, choose a group and a field theme, and prepare your Agentic AI proposal.
COMPETITION RULES / 競賽辦法