IdeateBuildPitchAI for Airports50 Students10 Teams20+ HoursAED 100K Prize PoolC-Level FinalsIdeateBuildPitchAI for Airports50 Students10 Teams20+ HoursAED 100K Prize PoolC-Level Finals
01 / The Mission
Real airport problems. Fresh minds. Working solutions.
The Dubai Airports Hackathon is a three-day innovation programme that challenges students to develop creative and impactful solutions to real-world opportunities and challenges at one of the world's busiest international airports.
// Objective
Give participants an engaging platform to collaborate, learn, innovate and present ideas with the potential to create meaningful impact.
// The format
Teams of four to six get at least 20 hours to transform an idea into a working or demonstrable solution, then pitch it to Dubai Airports Technology and C-level leadership.
0Students
0Teams
0+Build hours
0Days
0KAED prize pool
02 / Flight Plan
Three days from idea to impact.
Check the departures board. Tap any row for details.
Departures · HACK26DXB --:--:--
FlightTimeDateDestinationStatus
Ideate & CreateCheck-in and induction led by the Dubai Airports Technology team: an overview of DA Technology, its products and services, and the challenge briefing. Then teams start building. Runs 3:00 PM to 9:00 PM.Boarding
Develop & RefineA full build day from 9:00 AM to 9:00 PM. Teams complete and finalise their solutions with support from mentors.Scheduled
Technology Panel PitchesAll ten teams present to the Dubai Airports Technology Judging Panel, which evaluates every solution and shortlists the finalists.Shortlist
Final RehearsalShortlisted teams may arrive from 9:00 AM to practise their presentations. Solutions are locked: no further changes can be made.Code freeze
C-Level Finals & AwardsFinalists present to the C-level judging panel. Final judging is followed by the awards ceremony, which closes by 3:00 PM.Final call
Gate: DA Learning & Development CentreAll times GST (UTC+4)
01
Fri 13 Nov · 3 PM to 9 PM
Ideate & Create
Induction, challenge briefing and team kick-off. Participants begin developing their ideas and working together as teams.
02
Sat 14 Nov · 9 AM to 9 PM
Develop & Shortlist
Finish and polish your solution, then pitch to the Dubai Airports Technology team from 6:00 PM for evaluation and shortlisting.
03
Sun 15 Nov · 9 AM to 3 PM
Final Presentations
Rehearse from 9:00 AM. At 12:00 PM, finalists present to the C-level judging panel, followed by the awards ceremony.
03 / The Challenges
Pick your gate. Build the future.
Five real challenges from Dubai International Airport (DXB). Each team picks one and builds with synthetic data we supply, so no live system access is needed. Bring the ideas; we bring the data.
Gate A1Airport Operations · Featured
Universities
Disruption & Emergency Management
at Dubai International Airport (DXB)
When fog rolls in, a runway closes or an emergency unfolds, a small problem can quickly delay flights everywhere. Build a simple tool that helps the airport team spot a problem early, see who and what it affects, and decide what to do next. Pick any two situations, from bad weather and runway closures to aircraft emergencies and terminal evacuations.
What you get
Four APIs streaming synthetic real-time airport data. Every solution must show its information on the DXB airport map.
Flight Data APISurface Movements (CAT 11)Resource AllocationMETAR WeatherDXB Airport Map
Flight status, directions and airport services live in different places, so passengers have to search and piece things together. Build an AI assistant that helps passengers understand their journey, navigate the airport and get timely, personalised guidance, including in their own language and for accessibility needs.
What you get
A dev Flight Information API, airport mapping, approved Dubai Airports website content and synthetic passenger profiles.
Flight Information APIEsri MapsWebsite contentMultilingualAccessibility
A broken escalator or a planned lift shutdown near a gate can easily be missed until passengers are already arriving. Build an AI assistant that links arriving flights to the condition of the escalators, lifts and HVAC at their gates, and alerts the operations team before passengers are affected.
What you get
Flight data, a gate and stand mapping, and a synthetic asset register, work orders and permits to work for a sample concourse.
Flight dataAsset registerWork ordersPermits to workEsri Maps
Development information sits across many systems and is hard to turn into a clear plan. Design an AI advisor that compares an employee's evidence with their role and future-role requirements, then builds a practical, explainable development plan, while employees and line managers stay in control.
What you work with
One synthetic Excel workbook: 62 employees across 19 roles, a 14-competency framework, the Learning Hub catalogue, workplace activities and 15 test scenarios.
Reconciling company and customer ledgers is slow, manual and spreadsheet-heavy: formats differ, payments are partial or consolidated, and explanations sit in emails. Build a Finance assistant that standardises both ledgers, matches transactions on several fields at once, explains the exceptions and routes them through a controlled, auditable review.
Data and scenarios
Made-up company and customer ledgers, reference information and supporting emails only, with nine demo scenarios from a perfect match to an email-explained difference.
Multi-field matchingPartial and consolidated paymentsException explanationsFinance approvalAudit trail
Thank you to everyone who signed up. Registered teams, we will see you on Friday 13 November at the DA Learning & Development Centre, with check-in from 2:30 PM.
Your challenge: build a simple tool that helps the airport team spot a problem early, see who and what it affects, and decide what to do next. You will use synthetic (made-up) data only, so you can focus on the idea, not on access to real systems.
1. The problem
DXB is one of the busiest international airports in the world. Thousands of passengers connect between flights every hour, so a small problem in one place can quickly cause delays everywhere.
When something goes wrong, such as fog, a closed runway or an aircraft emergency, the information the team needs is spread across many systems and many organisations. This may cause four main pain points:
Late warning. Problems are often noticed only once flights are already delayed.
No single picture. Flight, weather, gate, passenger and aircraft ground movement information sit in different places, so it is hard to see what is happening on the airfield right now.
Hard to measure impact. It is slow to work out how many flights and passengers are affected.
Slow coordination. Airlines, ground teams and emergency services are updated by phone and radio, one by one.
2. What your solution should address
Pick any two of the situations below from disruption and emergency situation.
Situation
In simple terms
Disruption: Bad weather (fog, sandstorm, heavy rain)
Fewer flights can land and take off per hour, so delays build up.
Disruption: Runway closure
One of the two runways closes for a few hours.
Disruption: Gate or stand not available
A parking position or gate breaks, so aircraft and passengers must be moved.
Disruption: System failure
For example the baggage system stops and bags pile up.
Emergency: Aircraft emergency
A plane reports a technical problem and emergency services must be ready.
Emergency: Terminal evacuation
A fire alarm means passengers must leave part of the terminal.
3. The data and APIs you will be given
Dubai Airports will give every team access to four APIs. The data is synthetic data streamed in real time: it behaves like a live airport, but it does not represent real DXB flights, aircraft, passengers or weather.
API
What it gives you
Use it to
1. Flight Data API
Arrivals and departures: flight number, airline, origin / destination, scheduled and estimated times, status, runway, aircraft type and registration, passenger counts
Know which flights are affected and by how much
2. Surface Movements API
Live positions of aircraft and vehicles moving on the ground, based on ASTERIX CAT 11 data (see below)
Show aircraft and vehicles moving on the DXB map and spot problems on the airfield
3. Resource Allocation API
Which stand, gate, baggage belt and check-in counters each flight uses, and whether each resource is open or closed
Find knock-on effects and suggest re-allocations
4. Weather (METAR) API
Airport weather reports for OMDB: wind, visibility, runway visual range, weather (fog, dust, rain), cloud, temperature and pressure
Spot bad weather early and link it to delays
API documentation and access details will be shared by the organising team. Use the APIs as your main data source.
Surface movement data (ASTERIX CAT 11)
ASTERIX is the standard format used by air traffic systems to share surveillance data. Category 11 (CAT 11) carries the tracks of aircraft and vehicles moving on the airport surface, as seen by the airport's ground surveillance system (A-SMGCS). It is what controllers use to see the airfield, especially in fog or at night.
The Surface Movements API gives you this data as simple fields. The most useful ones are:
Field
Meaning
Example
track_number
Unique ID for each moving object
1042
time
Time of the position update
14:10:05
latitude, longitude
Position on the airport (WGS-84)
25.2528, 55.3644
callsign / registration
Who the target is (flight callsign or vehicle ID)
UAE123 / A6-EVA
target_type
Aircraft or vehicle (for example fire truck, follow-me, tow tug)
Aircraft
speed_kt, heading_deg
How fast and in which direction it is moving
15 kt, 120°
mode_3a / mode_s_address
Transponder codes linking the track to a flight
2341 / 896123
track_status
Moving, stopped, or lost
Moving
alert
Any safety alert linked to the track (for example runway incursion)
None
What you can do with it: draw aircraft and vehicles live on the DXB map, show queues of aircraft waiting at runway holding points, see which stands and taxiways are blocked, follow rescue vehicles during an emergency, and spot aircraft that have stopped where they should not be.
Weather data (METAR)
A METAR is the standard weather report for an airport, issued regularly (for example every 30 minutes) and used by pilots and controllers. The Weather API gives you METAR reports for OMDB (DXB), as the raw report and as simple fields.
Example (synthetic) report during fog: METAR OMDB 050000Z 12004KT 0200 R12L/0350N FG VV001 22/21 Q1012 NOSIG
Field
Meaning
From the example
station
Airport ICAO code
OMDB
time
Report time in UTC (Z). DXB local time is UTC+4
050000Z = 04:00 local on the 5th
wind_dir_deg, wind_speed_kt
Wind direction and speed (gusts shown with G)
120° at 4 kt
visibility_m
How far you can see, in metres
200 m
rvr_m
Runway visual range: visibility along a specific runway
Runway 12L: 350 m
weather
Codes such as FG fog, BR mist, DU dust, SA sand, RA rain, TS thunderstorm
FG (fog)
cloud / vertical_visibility
Cloud cover and height, or vertical visibility in fog
VV001 = 100 ft
temp_c, dewpoint_c
Temperature and dew point. When they are close, fog is likely
22 °C / 21 °C
qnh_hpa
Air pressure
1012 hPa
trend
Expected change in the next 2 hours
NOSIG (no big change)
What you can do with it: warn when visibility or RVR is falling, predict fog from a closing temperature and dew point gap, flag strong winds or dust, and show the current weather next to your map.
Extra data you can create yourself
If your idea needs it, you can add small datasets of your own in a spreadsheet, CSV or JSON file:
All solutions must use the Dubai International Airport (DXB) map as the base for showing aircraft, vehicles, stands and gates. You can choose either option:
Option 1: Image. Use an image of the DXB airport layout as a background and place your data on top of it.
Option 2: Official AIP chart. Use the aerodrome charts for OMDB (the ICAO code for DXB) published in the UAE Aeronautical Information Publication (AIP) by the General Civil Aviation Authority: UAE eAIP (GCAA website) (see section AD 2 OMDB).
Tip: the surface movement data uses latitude and longitude, so line your map up with real coordinates (georeference it). A good check is the airport reference point published in the AIP: 25°15'10"N, 55°21'52"E. If the map is not aligned, aircraft will appear off the taxiways.
5. Scenarios to recreate with the synthetic data
Use two or more of these as your demo story. Each one tells you what to simulate and what your tool should show.
Scenario
What happens (simulate this)
What your tool should show
A. Morning fog
Visibility drops to 200 m from 04:00 to 08:00, shown in the METAR reports. Only about half the normal flights can land each hour.
A warning before it starts, which flights will be delayed, which passengers may miss connections, and aircraft taxiing slowly on the map.
B. Runway closure
At 14:00 one runway closes for 2 hours because of debris.
How many flights and passengers are affected, aircraft queuing at the open runway on the map, and a suggested new departure order.
C. Aircraft emergency
An arriving aircraft with 300 passengers reports an engine problem 20 minutes before landing.
An alert to the right teams, rescue vehicles moving on the map, a timeline of actions, and the impact on other arriving flights.
D. Terminal evacuation
A fire alarm at 19:30 clears four gates in Concourse B for 45 minutes.
Which flights and passengers are affected, and new gate assignments after the all clear.
6. What success looks like
From the airport's point of view, a great solution:
Warns early: raises an alert before the problem causes delays.
Shows one clear picture: a simple dashboard with the DXB map showing aircraft and vehicles moving live, which anyone can understand in seconds.
Measures impact: shows the number of flights, passengers and connections affected.
Suggests actions: gives clear options (for example delay, change gate, divert) and explains why.
Keeps everyone informed: shows who needs to be notified and what message they get.
Judges will value a clear, working demo and a simple user experience over a complex design.
7. Rules to keep in mind
Safety first. Your tool should never suggest anything that puts safety at risk to save time.
People decide, the tool advises. Recommendations support the airport team; a person always makes the final call.
Air traffic control manages the runways, and emergency services lead in an emergency. Your tool helps coordinate them, it does not replace them.
Big aircraft need big gates. An aircraft like the A380 can only use gates built for it.
Recovery takes time. When a problem ends, delays do not disappear instantly; flights clear gradually.
Use the DXB map. Every solution must show its information on the DXB airport map (image or AIP chart).
Synthetic data only. All API data is synthetic data streamed in real time and does not represent the real airport. Do not use real passenger names or personal information.
Your challenge: build an AI-powered assistant that helps passengers understand their journey, navigate the airport and receive timely, personalised guidance.
1. The problem
Passengers often need information from several sources before they can confidently decide what to do next. Flight status, terminal directions, airport services and website content may be available separately, but the passengers must still search, interpret and combine them.
Fragmented information. Journey details, directions and airport services are spread across multiple channels.
Generic guidance. Available information may not reflect the passenger’s flight, location, time or immediate need.
Navigation complexity. Passengers may struggle to locate gates, lounges, shops, transport and essential services.
Experience gaps. Language, accessibility and unfamiliarity with the airport can make simple journeys stressful.
2. What your solution should address
Create a simple passenger-facing intelligence assistant that combines trusted information and provides clear answers, recommendations and navigation support. The solution should cover at least three of the following use cases:
Use case
In simple terms
Flight journey assistant
Explain the passenger’s current flight status, key milestones and what action to take next.
Airport navigation
Guide the passengers to gates, lounges, retail, food, transport and airport services using the airport map.
Personalised recommendations
Suggest relevant services or experiences based on available time, terminal, preferences and journey stage.
Frequently asked questions
Answer questions using approved Dubai Airports website content and challenge data.
Accessible and multilingual support
Offer concise guidance suitable for different languages and accessibility needs.
3. The data and APIs you will be given
Teams may use the following data sources and tools to build the assistant.
Source / tool
What it may provide
Possible use
Flight Information API
Dev flight information API for arrivals and departures, flight status, scheduled and estimated times, terminal, concourse and gate. Note: only publicly available flight information is included; no passenger or confidential operational data is exposed.
Give flight-specific updates, connection guidance and proactive alerts.
Esri Maps SDK / mapping service
Airport map layers, points of interest, routing and location context, subject to final validation.
Show routes and guide passengers to airport facilities and journey touchpoints.
Dubai Airports website content
Approved public information collected from selected pages, subject to technical and legal validation.
Answer FAQs about airport services, facilities, transport and passenger processes.
Additional validated datasets
Services, amenities, operating hours, accessibility details, retail or journey content where available.
Improve recommendations and make responses more contextual.
Synthetic passenger context
Synthetic flight, language, preferences, mobility needs and journey stage.
Demonstrate personalisation without using real passenger information.
4. Synthetic Passenger data
All passenger and behavioural information used in the demo must be synthetic. Teams must not use real passenger names, contact details, booking references, travel documents or personally identifiable information.
Use three or more of these as your demo story. Each one tells you what the assistant should show.
Scenario
What happens
What the assistant should show
A. Smooth departure
A passenger arrives three hours before departure and wants to know the best sequence of steps.
A personalised journey plan, navigation to key locations and timely reminders.
B. Gate change
The gate changes shortly before boarding.
A clear alert, updated walking route, estimated walk time and impact on planned activities.
C. Long dwell time
A passenger has several hours before departure or connection.
Suitable dining, retail, lounge or experience options based on location and available time.
D. Passenger FAQ
The passenger asks about transport, baggage, Wi-Fi, accessibility or airport services.
A concise answer grounded in approved website content, with the relevant next step.
E. Accessible and multilingual journey
A passenger who prefers Arabic and uses a wheelchair asks for help reaching their gate.
Concise guidance in the passenger’s preferred language, a step-free route with lifts and assistance points, and a realistic, estimated travel time.
6. What success looks like
From the airport's point of view, a great solution:
Context-aware guidance. Understands the passenger context and gives relevant, timely guidance.
One joined-up experience. Combines flight information, mapping and approved airport content into one clear experience.
Clear navigation. Provides map-based navigation or a clear route to the next journey touchpoint.
Safe handling of uncertainty. Handles change and uncertainty safely, without presenting unsupported information as fact.
Measurable performance. Demonstrates measurable performance, including response latency and successful task completion.
Judges will value a clear, working demo and a simple user experience over a complex design.
7. Rules to keep in mind
Passenger safety comes first. The assistant must never recommend bypassing airport, airline, immigration, security or safety procedures.
Guidance, not instruction. The assistant provides guidance, while official operational instructions remain authoritative.
Synthetic data only. Use only synthetic passenger data and approved challenge content.
No commitment. Do not make commitments on behalf of airlines, government entities, ground handlers or airport operations.
Confirmed versus estimated. Clearly distinguish confirmed flight information from recommendations or estimated walking times.
Your challenge: build an AI-powered assistant that links arriving flights to the condition of the airport assets at their gates and alerts the operations team before passengers are affected.
1. The problem
Before a flight arrives, the operations team needs to know whether the gate area is ready to receive passengers. Flight schedules, asset status, maintenance work orders and permits to work are held in separate systems, so a broken escalator or a planned lift shutdown near an arriving flight’s gate can easily be missed.
Disconnected data. Flight, asset, work order and permit information sit in different systems.
Late discovery. Asset faults near a gate are often noticed only when passengers are already arriving.
Missed planned work. An active permit to work on an escalator or lift may not be linked to the flights using that area.
Passenger impact. Out-of-service assets can delay disembarkation and boarding, or force passengers onto longer routes.
2. What your solution should address
Create a simple operations-facing assistant that combines flight and asset information and raises clear, early alerts. The solution should cover at least three of the following use cases:
Use case
In simple terms
Flight-to-gate asset check
For each flight arriving in the next hour, identify its gate and the assets serving that gate, such as escalators, lifts and HVAC.
Asset disruption alert
Alert the operations team when an asset serving an arriving flight’s gate is out of service or has an open incident.
Planned work awareness
Flag active or upcoming permits to work and maintenance on assets near gates with arriving flights.
Impact and recommendation
Explain the likely impact of the passengers and suggest an action, such as moving the flight to another gate or using an alternative route.
Operations summary
Show upcoming flights, affected gates and recommended actions in a simple message, dashboard or chat view.
3. The data and APIs you will be given
Teams may use the following data sources and tools. Assets, work order and permit data are synthetic and prepared for a sample concourse, so that flights, gates and assets match. Teams who can generate or source their own realistic data are welcome to do so, as long as it stays consistent and contains no confidential information.
Source / tool
What it may provide
Possible use
Flight data
Dev flight information API arrival schedule, refreshed hourly, with flight number, scheduled and estimated arrival time, concourse and gate. Note: a dev flight information API containing only publicly available flight information may also be provided for reference.
Identify which flights arrive at which gates in the next hour.
Gate–stand mapping
Reference table linking each aircraft stand to its concourse and the passenger gate(s) it serves. Arrival flights in flight API include the stand only, not the gate.
Convert each arriving flight’s stand to a gate, then find the assets serving that gate.
Asset register
Synthetic list of escalators, lifts and HVAC units for a sample concourse, with location, zone and gates served. Based on a simplified asset management structure.
Link each gate to the assets passengers rely on.
Work orders and incidents
Synthetic, simplified work order records showing asset, type, status, reported time and expected completion.
Detect assets that are broken down or under repair.
Permits to work (PTW)
Synthetic permit records showing assets, work description, start and end time and status.
Flag planned work that overlaps with arriving flights.
Esri Maps SDK / mapping service
Concourse map layers, gate and asset locations, subject to final validation.
Show affected gates, out-of-service assets and alternative routes on a map.
4. Synthetic data
All assets, work order and permit data used in the demo are synthetic and do not represent real DXB operations. The datasets are linked by gate and asset ID, so a flight, the assets at its gate and any open work can be matched.
13:10, DA215, C12, escalator and lift out of service, consider gate C9
5. Scenarios to recreate
Use three or more of these as your demo story. Each one tells you what the assistant should show.
Scenario
What happens
What the assistant should show
A. Gate ready
A flight is due at gate C7 in one hour and all assets serving the gate are working.
A “ready” status for the gate, with no action needed.
B. Escalator breakdown
An open breakdown work order exists for the escalator serving gate C12, and a flight is due there in 45 minutes.
An alert to the operations team names the flight, gate and asset, with the likely passenger impact.
C. Planned work overlap
A permit to work is active on a lift near gate C10 during a flight’s arrival window.
A warning that planned work overlaps the arrival, showing the permit time window.
D. Multiple issues and reroute
Several assets in one zone are out of service, and three flights are due there within the hour.
A prioritised list of affected flights and a recommendation to move one or more to gates with working assets.
E. Step-free access risk
The only lift serving gate C5 is out of service while a flight is due there.
An alert highlighting the step-free access risk and suggesting an alternative gate or route.
6. What success looks like
From the airport's point of view, a great solution:
Early alerts. Alert the operations team early enough to act, for example one hour before arrival.
One joined-up view. Correctly links flights, gates, assets, work orders and permits.
Clear recommendation. Explain the impact and suggest a practical action, such as a gate change or alternative route.
Safe handling of uncertainty. Distinguishes confirmed asset status from estimates, without presenting assumptions as fact.
Simple delivery. Presents alerts in a message, dashboard or chat that an operations user can understand quickly.
Judges will value a clear, working demo and a simple user experience over a complex design.
7. Rules to keep in mind
Safety comes first. The assistant must never recommend passing safety procedures, permits to work or equipment isolation controls.
Recommendation, not instruction. The assistant supports the operations team; gate allocation and maintenance decisions remain with authorised staff.
Synthetic or public data only. Use synthetic challenge data, data you generate yourself or public information. Do not use real confidential operational data.
No commitment. Do not make commitments on behalf of airlines, ground handlers, maintenance contractors or airport operations.
Confirmed versus estimated. Clearly distinguish confirmed asset status and flight times from estimated impact or recommendations.
How might we use AI to help Dubai Airports employees understand their development opportunities and build practical, evidence-based development plans, while keeping employees and line managers in control?
Your challenge: Design and prototype an AI Professional Development Advisor that provides personalised, evidence-based and actionable development recommendations. You will use synthetic (made-up) data only, so you can focus on the idea, not on access to real systems.
1. The problem
DA Employees and Line Managers often have access to a significant amount of development information, but that information exists across different systems and is difficult to translate into a clear development plan.
For example, development decisions may need to consider:
What competencies are required for an employee's current role?
What does their performance review indicate about strengths and development areas?
What learning has the employee already completed?
What capabilities would they need for a potential future role?
Which learning resources are relevant to the identified gaps?
Would formal learning actually be the best intervention, or would coaching, mentoring, job shadowing, stretch assignments or project exposure be more appropriate?
Your Advisor brings these sources together: role requirements, performance evidence, career goals, learning history and the Learning Hub catalogue.
2. What your solution should address
Your AI Agent should be able to complete the following journey:
Step
Action
Understand the employee
Identify current role, required competencies and proficiency levels, available performance evidence, career goal, target role where applicable, and completed learning.
Identify current-role competency gaps
Compare current-role requirements vs available evidence about the employee.
Identify competencies currently meeting expectations, competency gaps, development areas, existing strengths that could develop further.
Explain why it identified a gap rather than simply returning a score.
Explore future-role readiness
Compare the employee's available capability evidence with the requirements of the target role identified in their career goal. Clearly separate current-role performance from future-role development.
Recommend personalised development
Recommend a simple combination of relevant Learning Hub content and practical development, such as coaching, job shadowing, peer learning or a stretch assignment. Check learning history to avoid unnecessary repetition.
Create an actionable development plan
Organise recommendations into a practical sequence: Competency gap or strength -> Development objective -> Recommended resource or activity -> Expected outcome -> Suggested timeframe.
Include Line Manager approval
Activities that take time away from the role, involve other people's time or cost money (coaching, mentoring, job shadowing, stretch assignments, projects, workshops and programmes) need Line Manager approval. Self-paced e-learning, microlearning, reading and peer learning do not; the dataset marks every item. A possible workflow:
Employee asks Agent → Agent analyses → Agent recommends → Employee reviews → Manager approves → Development plan agreed → Actions begin
3. The data you will need
All the data is in one Excel workbook of synthetic data covering 62 employees and 19 roles. Convert it to CSV or JSON if that suits your tool. Start with the README sheet: it explains how the sheets connect and the rules for missing, conflicting and outdated evidence.
Dataset
Main fields
Example
Employee profiles
Employee ID, name, department, current role, years in role, line manager, career goal and target role.
EMP002 is a Team Supervisor who wants to become an Operations Manager.
Role requirements
Each role's required competencies and required level (1 to 4), with a short level description.
The Operations Manager role requires Strategic Thinking Level 3.
Competency framework
The 14 competencies with a description of each level from 1 (Basic) to 4 (Advanced).
Data Analysis Level 3 means "Independently analyses trends and explains findings."
Career paths
Typical moves between roles and the usual time in role before a move.
Team Supervisor to Operations Manager after 2+ years.
Employee evidence
Observed competency level, evidence source, evidence date and evidence summary.
EMP001 demonstrates Data Analysis Level 2 based on a performance review.
Development resources
Workplace and social learning: coaching, mentoring, job shadowing, stretch assignments, projects and peer learning, with the level step and approval need.
Operational Data Mini-Project requires Line Manager approval.
Learning Hub catalogue
Formal learning: title, competency, level step, format, duration, provider, approval need, mandatory roles and link.
Interpreting Operational Data is a two-hour e-learning for progression from Level 2 to Level 3.
Learning history
Completed, in-progress and assigned learning, with date, score and how the learning was applied.
EMP002 has completed Strategic Thinking Foundations.
Test and validation data
15 scenario prompts with expected results, suggested recommendations, approval needs and prohibited outputs.
S1 to S3 are the core demo below; S4 to S15 test harder cases such as missing evidence and privacy.
4. Scenarios to recreate with synthetic data
Demonstrate your Agent with these three core scenarios. Then try some of the harder test scenarios in the dataset (S4 to S15), such as missing or conflicting evidence, a new joiner, overdue mandatory learning or a request to see a colleague's data.
Scenario
What Employee might ask
What your tool should show
1 - Current-role competency gap
"What should I develop for my current role?"
Data Analysis is observed at Level 2 while the Operations Analyst role requires Level 3. Explain the evidence. EMP001 has already completed Data Storytelling Microlearning, so recommend Interpreting Operational Data followed by the Operational Data Mini-Project. The mini-project requires Line Manager approval.
2 - Preparing for a future role
"What should I develop to become an Operations Manager?"
Strategic Thinking is observed at Level 2 while the target Operations Manager role requires Level 3. EMP002 has already completed Strategic Thinking Foundations, so recommend Thinking Beyond Daily Operations followed by job shadowing with an Operations Manager. Job shadowing requires Line Manager approval. State clearly that development supports readiness but does not guarantee promotion.
3 - Strong performer with no current-role gap
"I am doing well. What could I learn next?"
Learning Design exceeds the current-role requirement and Communication meets it. Do not invent a gap. Recommend the Learning Design Showcase as an enrichment activity; no Line Manager approval is required.
5. What success looks like
A successful prototype should demonstrate more than a chatbot interface:
1. Evidence-based
Recommendations can be traced back to the synthetic employee profile, role requirements, performance evidence, career goal or learning history.
2. Personalised
Two employees in the same role should not automatically receive identical recommendations.
3. Gap-aware
The Agent correctly distinguishes:
Current-role gaps
Capability enhancement
Future-role development
4. Actionable
Recommendations translate into practical activities rather than generic advice.
5. Learning ecosystem connected
Relevant Learning Hub resources are recommended against specific development needs.
6. Beyond learning courses
The Agent appropriately uses coaching, mentoring, job shadowing, projects, stretch assignments and other experiential approaches.
7. Explainable
The user can ask:"Why are you recommending this?" and receive a clear explanation.
8. Human-controlled
The system appropriately involves the Line Manager in reviewing or approving advanced development actions. Simple activities, such as reading recommended articles or completing self-paced modules, do not require Line Manager approval.
9. Safe and fair
Sensitive employment decisions are not made autonomously.
10. Handles uncertainty
The Agent identifies missing or contradictory evidence rather than hallucinating an answer.
Judges will value a clear, working demo and a simple user experience over a complex design.
6. Rules to keep in mind
Rule 1: AI advises, humans decide.
Your Agent is a Professional Development Advisor, not an automated talent decision-maker. It must not independently decide:
Whether someone should be promoted
Whether someone should receive a salary increase
Whether someone should enter succession planning
Whether someone should receive a poor performance outcome
Whether disciplinary action should be taken
The Advisor complements the Line Manager's judgement; it does not replace it.
Rule 2: Recommendations must be explainable
The Agent should be capable of showing: Evidence → Gap → Recommendation rather than: Employee data → Black box → Recommendation
Rule 3: Do not create information that does not exist
If evidence is missing, the Agent should say so. "Insufficient evidence" is an acceptable answer.
Rule 4: Do not treat development needs as negative labels
Use development-focused language. Prefer: "Development opportunity", "Capability to strengthen", "Future-role development area". Avoid unnecessarily labelling people as weak, poor performers or unsuitable.
Rule 5: Respect privacy and access
The prototype should follow the principle: Only use information the user is authorised to access.
Rule 6: Separate current performance from future potential
A person can perform extremely well in their current role while still having significant development needs for a different future role. The Agent should not confuse: Current performance ≠ Future-role readiness
Rule 7: Avoid guaranteed career outcomes
The Agent may say: "Developing these capabilities could help prepare you for the Operations Manager role."
It should not say: "Complete these activities and you will be promoted."
Rule 8: Think beyond training
Not every competency gap needs a course. Students should consider a 70-20-10 style mix of experiential, social and formal learning where appropriate, rather than automatically mapping every gap to e-learning.
Build more than a course recommendation tool. Create an AI-powered development advisor that understands the employee’s context, compares available evidence with role requirements, and recommends an appropriate combination of learning, workplace experience, and support.
The solution should explain why each recommendation has been made, identify when evidence is missing or contradictory, protect employee privacy, and keep the employee and Line Manager in control.
AI should support better development conversations, not make employment decisions.
Your challenge: build an AI-enabled reconciliation solution that compares company and customer ledgers, intelligently matches transactions, identifies and explains exceptions, uses supporting information where relevant, and provides Finance with a controlled, auditable workflow for investigation and resolution.
1. The problem
Finance teams reconcile company and customer ledgers through a sequence of data collection, transaction comparison, exception investigation, agreement of corrective actions, closure and documentation. The process is often manual, fragmented and reactive.
Manual effort. Transaction matching is time-consuming and depends heavily on spreadsheets and individual review.
Inconsistent data. Invoice, payment and reference formats differ, and customers may provide an incomplete ledger statement.
Complex matching. Timing differences, partial payments, consolidated payments, payments in transit make direct matching difficult and unrecorded document.
Data quality issues. Transactions can be missing, duplicated, incorrectly recorded or mapped to the wrong account.
Fragmented evidence. Supporting explanations may sit separately in emails or other records.
Limited control visibility. Aging, ownership, status and resolution of exceptions are not always easy to monitor.
2. What your solution should address
Create a simple Finance-facing assistant that standardises two ledgers, performs intelligent multi-field matching and routes unresolved items through a controlled review workflow. The solution should cover at least three of the following use cases:
Use case
In simple terms
Ledger ingestion and standardisation
Load company and customer ledgers, map fields and normalise dates, amounts, references and descriptions.
Intelligent transaction matching
Match using combinations of amount, invoice number, date, customer reference and description rather than one field only.
Complex payment matching
Support one-to-one, one-to-many and many-to-one matching, including partial and consolidated payments.
Exception detection and explanation
Identify missing items, amount or date differences, duplicates, unapplied payments and mapping issues, then suggest a clear explanation.
Controlled investigation workflow
Show confidence, financial exposure, owner, status, evidence and Finance approval or rejection of AI suggestions.
Management reporting
Summarise total transactions, auto-matched items, suggested matches, exceptions, items under investigation and resolved items.
3. Synthetic data
All ledger and supporting information used in the demo must be synthetic and must not represent real customer or Dubai Airports financial data. The datasets can be linked through customer, invoice, payment, document and accounting references.
Important: The matching engine should not depend on one field only. It should evaluate combinations such as amount + invoice number + date + customer reference + description.
4. Typical exceptions
Missing transaction on either ledger
Amount mismatch for the same transaction
Timing difference between posting dates
Reference mismatch despite otherwise similar transactions
Partial payment
One-to-many or many-to-one matching
Duplicate invoice or payment
Unapplied or unidentified payment
Missing or mismatched credit note or adjustment
Incorrect account or transaction mapping
5. Scenarios to recreate
Use three or more scenarios as the demo story. Each scenario should show the source records, AI result, confidence or rationale, and the Finance review outcome.
Scenario
Company ledger
Customer ledger
What the assistant should show
A. Perfect match
TR001, AED 100,000
TR001, AED 100,000
Auto-match with clear matching rationale.
B. Amount difference
TR 002, AED 75,000
TR 002, AED 72,500
Flag an AED 2,500 variance and route for review.
C. Missing invoice/ transaction
TR 003, AED 50,000
No entry
Flag a missing transaction on the customer ledger.
D. Reference difference
TR 004, AED 30,000
Ref. 004-A, AED 30,000
Suggest a probable match using amount, date and description.
E. Partial payment
TR 005, AED 100,000
Payment AED 60,000
Match partially and show AED 40,000 outstanding.
F. Consolidated payment
Three invoices TR 006, TR007 & I TR008 totaling AED 150,000
One payment AED 150,000
Identify a many-to-one match and list linked invoices.
G. Duplicate
TR009 appears once
TR009 appears twice
Flag a possible duplicate and prevent forced clearance.
H. Timing difference
Posted 31-Oct
Posted 1-Nov
Identify a likely timing difference within an approved tolerance.
I. Email-supported exception
Invoice AED 90,000
Customer shows AED 85,000
Use relevant synthetic email context to explain the AED 5,000 difference.
6. What success looks like
From Finance’s point of view, a great solution:
Joined-up processing. Ingests and standardises both ledgers before comparison.
Accurate matching. Automatically matches clear transactions using multiple fields and supports complex match relationships.
Useful exceptions. Categorises mismatches, quantifies unresolved financial value and explains likely causes.
Transparent confidence. Shows a confidence score and matching rationale for suggested matches.
Human control. Allows Finance to review, approve or reject AI suggestions and assign exception ownership.
Auditable evidence. Retains source records, matching logic, Finance decisions, status changes and supporting evidence.
Clear reporting. Presents total transactions → auto-matched → suggested matches → exceptions → under investigation → resolved.
Simple delivery. Provides a clear dashboard, message or chat experience that Finance users can understand quickly.
Judging emphasis: Judges should value a clear, working demo, sound financial controls and a simple user experience over unnecessary technical complexity.
7. Rules and controls to keep in mind
No forced matching. Transactions must not be cleared only because the AI finds a reasonably similar item.
Controlled tolerances. Amount, date and reference tolerances may be configurable, but Finance must approve production thresholds.
Human review for uncertainty. High-confidence exact matches may be proposed for automated treatment; lower-confidence or material differences must go to Finance review.
Segregation of duties. Preparation, review and approval responsibilities must remain appropriately separated.
Evidence before closure. An exception is not resolved merely because AI proposes an explanation; appropriate support is required.
Full audit trail. Retain source records, matching logic and result, Finance decision, exception status and supporting evidence.
Synthetic or public data only. Do not use real confidential customer or operational financial data in the hackathon demo.
Recommendation, not accounting approval. The assistant supports Finance; authorised staff retain financial judgment and approval.
HACKATHON OBJECTIVE Build an AI-enabled reconciliation solution that compares company and customer ledgers, intelligently matches transactions, identifies and suggests explanations for exceptions, uses supporting information where relevant, and provides Finance with a controlled, auditable workflow for investigation and resolution.
Design principle: Automate matching, accelerate investigation, but keep financial judgment and approval with Finance.