Tech
Galaxy
Award
tga.org
Cohort logistics
TGA is built around small, intentional cohorts. The numbers below are not negotiable — they are the conditions under which our seven-day pedagogy works.
- Eligibility: Secondary school & university students
- Team size: 8 students per team
- Language: English (working proficiency)
- Host location: Singapore · Tier-1 university campus
- Format: Immersive · project-based
- Faculty: NUS · NTU · HKU · Cambridge
- Cohort size: Capped at 12 teams per year
- Annual frequency: Once per summer
Five challenge components
Each cohort moves through five sequential components. Each is a gate: no team advances to design before research is in place, and no team advances to defense before its design is documented. This sequence is deliberately mimicked from how real academic and policy work is conducted.
01
Problem identification
Teams choose a real urban issue, scoped specifically enough that it could plausibly be addressed within seven days. Faculty veto vague framings on day one.
02
Research & framing
Teams draw from public datasets, urban fieldwork in Singapore, and expert talks. The deliverable is a defensible problem statement, not a solution.
03
AI application
Teams select an AI concept appropriate to their problem and explain — in their own words, without slide-deck buzzwords — why it fits and what it costs.
04
Design solution
Teams visualize the idea using models, maps, journey diagrams, or working prototypes. The output must be legible to a stranger on a poster wall.
05
Presentation & defense
A 10-minute live pitch followed by 3 minutes of panel Q&A. The defense is the harder half — judges probe assumptions, not summaries.
Daily schedule
Seven days, paced so that thinking time and pressure time are both protected. Mornings are for input and exploration; afternoons are for synthesis; evenings are for written reflection. The schedule below is representative — exact timings vary by cohort.
Cohort welcome, campus tour, opening keynote from faculty. Teams begin scoping their urban problem. Faculty mentors meet teams one-on-one.
Site visits relevant to each team's problem — public transit nodes, housing precincts, civic spaces. Field notes due that evening.
Public-dataset workshop. Expert salons with urban planners, AI researchers, and policymakers. Teams produce a written problem statement.
Teams select their AI approach and present a five-minute reasoning to mentors. Mentor pushback is structured but firm.
Visualization and prototyping intensive. Teams build the artifacts they will defend: models, maps, mockups, or working demos.
Full rehearsal before mentor panel. Hard feedback. Teams revise overnight.
10-minute pitch plus 3-minute Q&A before the full judging panel. Galaxy awards announced that evening.
Judging framework
Every project is scored against a published rubric. The rubric is shared with teams on day one and made public after each cohort. Honors are decided by the full judging panel; each panelist's scoring is documented and audited.
Judging panel
Professors from the National University of Singapore (NUS), Nanyang Technological University (NTU), the University of Hong Kong (HKU), and the University of Cambridge. Guest panelists from partner research and urban-design institutions.
Rubric dimensions
- AI logic & technical soundness 25%
- Urban systems understanding 20%
- Innovation & creativity 20%
- Design communication & clarity 15%
- Defense rigor & responsiveness 10%
- Team integration & role equity 10%
What makes
this hard.
TGA is not a STEM fair. The work expected of a TGA participant approximates the work expected of a first-year graduate student at a serious policy school. We say this plainly because applicants should know what they are agreeing to.
This is, at TGA:
— An AI-integrated urban systems design challenge
— Where solutionism is not rewarded, but systems complexity is
— Where students defend their assumptions and reflect on social impacts
— Where visual clarity must complement logical depth
Students must, at TGA:
— Work across five to six knowledge domains
— Handle unstructured ambiguity, not multiple-choice prompts
— Translate abstract concepts into audience-centered presentation
— Survive a real academic Q&A, on the record
The 2026 brief cycle.
The seven days in Singapore are the visible week. The cycle is the year around them. 2026 is the first year we are publishing the cycle in this level of detail, so that a teacher can place TGA against SSEF, ISEF Phoenix (9–15 May), the iGEM Jamboree in Paris (13–16 November), and the rest of the landscape on The Field.
T−10 months · Sep 2025
Expression of interest
Adult leads write. We reply with Stage 1. No brief yet — we are selecting teams, not solutions.
T−9 months · Oct 2025
Offers and confirmation
Twelve teams, two named alternates. Fee or bursary confirmation. Invitation letters for visas.
T−6 months · Jan 2026
Brief published
Confirmed teams receive the 2026 urban brief, the pre-reading list, the role sheet, and the rubric that will still be in force on day seven.
T−4 months · Mar 2026
Roles locked
Each team returns a role map of eight. Changes after this date require a reason. Shadows are not a role.
T−2 months · May 2026
Pre-reading complete
Annotated notes due from the adult lead. This is also ISEF week in Phoenix — do not confuse the two pieces of work.
T−0 · 6–12 Jul 2026
Singapore intensive
Seven days as scheduled below. Certificates issued at close. Public record updated within ten days.
The 2026 brief will name a small set of urban systems that Singapore makes unusually legible: public housing as a civic technology, transit nodes as social infrastructure, heat and water as design constraints, and the civic trust that any "smart" intervention either spends or deposits. Teams will not be assigned a system at random. They will choose, on day one, inside that set, and they will be asked to scope until a stranger could recognise the problem on a walk. Faculty vetoes are not a humiliation. They are how we keep the week from becoming a dashboard competition.
Pre-reading is short on purpose: a handful of public documents, a data primer, and two essays that disagree with each other. We are not running a correspondence course. We are making sure that day one does not start from zero. Teams that skip the reading arrive charming and lose Tuesday.
After July, the cycle has a quiet tail. Adult leads receive digital certificates. The public verifier is updated. A short internal audit of the rubric scores is filed. We do not publish individual score sheets. We do publish, each year, a one-page note if we change a rubric weight. That note is part of the archive, not a press release.
Eight people, eight jobs.
Roles are a design problem. They are also how we judge the Constellation award. The titles below are the 2026 default map. Teams may rename them. Teams may not collapse them into "the two who speak and the six who nod."
Framer
Owns the problem statement. Protects it from becoming a slogan. Speaks first on day one and last on day six if the statement has drifted.
Field lead
Owns the Singapore walks: where, why, what is written down. A team that never leaves the seminar room is not doing TGA, regardless of how pretty the slides are.
Evidence lead
Owns datasets, citations, and the difference between a number and a claim. Lies of precision are still lies.
AI reasoner
Owns the argument for the chosen AI concept: why it fits, what it costs, what it cannot see. Not "the coder," unless the coder can also explain.
Ethicist
Owns who is harmed if the idea works, and who is erased if it works as designed. Has veto power over a demo that is merely clever.
Prototype lead
Owns the artifact a stranger can stand in front of: model, map, mock, working slice. Beauty without legibility is a fail.
Defense captain
Owns the ten-minute structure and the three-minute triage. Does not own every answer. Redirects a question to the person who actually knows.
Integrator
Owns the joints: who is silent, which role is decorative, whether the eight still describe one project. Often the quietest person. Often the reason Constellation is possible.
The adult lead is not a ninth student. The adult lead is logistics, safeguarding, and the person who stops a team from pulling an all-nighter that wrecks day seven. Mentors will talk to students, not through the adult, unless safety requires it. Galaxy Mentor — the eighth honor — is conferred by anonymous student nomination, not by how often the adult spoke in the hall.
If a team arrives without a role map, we will make them write one before lunch on day one. If a team arrives with a role map that is fiction, the panel will notice by day six. Role equity is 10% of the published rubric because we have seen what happens when it is 0%.
Judging rubric, in force 2026.
Weights are unchanged from the list already on this page. What follows is how a panelist is asked to read them. Scores are independent; honors are discussed. Polaris requires unanimity. North Star does not.
AI logic & technical soundness — 25%
Did the team choose an AI concept that could actually touch the problem they framed? Can they say what data it would need, what it would fail on, and what a non-AI alternative would cost? A large language model used as a magic civic brain scores low. A smaller, boringer method that the team can defend scores high. Buzzwords are not evidence. A diagram of a pipeline is not evidence unless someone can walk it.
Urban systems understanding — 20%
Did they see a system, or a building? Can they name the public who would live with the intervention? Did the Singapore field notes change the problem statement, or were the notes tourism? Panelists who have walked the same precinct will know. This dimension is why we convene here and not on a video call.
Innovation & creativity — 20%
Not novelty for its own sake. The Supernova award lives here, but the rubric asks a prior question: is the surprise defensible? A team that reframes a tired "smart bin" brief into a question about labour and heat may outscore a team with a spectacular demo of the bin. Creativity that cannot survive "why should this exist?" is decoration.
Design communication & clarity — 15%
Could a stranger on the poster wall understand the idea in ninety seconds without a guide? Are the visual artifacts doing work, or covering the absence of an argument? Stardust lives near this dimension. So does ordinary professionalism. Tiny type, untranslated jargon, and a ten-minute talk that is actually seventeen minutes all cost points.
Defense rigor & responsiveness — 10%
The three minutes. Did they hear the question that was asked? Did they know when to say "we do not know"? Did they protect a weak claim or abandon it? A fluent non-answer scores below a clumsy true one. The panel will address at least one question to a student who has not yet spoken.
Team integration & role equity — 10%
Eight people, eight jobs, no shadow players. Mentors' notes from the week inform this score; the defense cannot fully hide a team that was two people all along. Constellation is the honor that sits here. A brilliant soloist in a quiet team can still win Rising Star as an individual. The team score will not pretend the other seven were essential if they were not.
Defense protocol.
Day seven is not a talent show. It is a short viva with an audience. The protocol below is what teams receive in January and what the panel uses in July. Clock time is merciless because mercy here would be unfair to the team that rehearsed.
- Order: Drawn the evening of day six; not negotiated
- Setup: 4 minutes, silent, team only
- Pitch: 10 minutes · lights warn at 9:00
- Q&A: 3 minutes · panel owns the floor
- Exit: 2 minutes to strike · next team in
- Language: Spoken English · bilingual artifacts allowed
- Recording: Internal archive · not a public livestream
- Objections: Adult lead, in writing, after the ceremony — not during Q&A
During the pitch, the panel does not interrupt except for a genuine technical failure. During Q&A, the panel interrupts freely. A student who attempts to run out the clock with a second speech will be stopped. A student who answers a question that was not asked will be asked again. A student who blames a missing teammate will not be helped by that sentence.
At least one question will go to a student who has not spoken. At least one question will ask about a cost, a public, or a failure mode. At least one question will be unkind in the academic sense: it will doubt a central assumption. That is the job. Mentors who coached the team during the week do not sit on that team's scoring panel. Guest panelists are introduced by name and affiliation at the start of the morning, not as a surprise at minute nine.
The Galaxy is announced the same evening. Panelists do not give private recaps to teams in the corridor. Adult leads who want a pedagogical debrief may request one in writing after the cohort; we will schedule it when we can. We will not reopen a honor because a parent is disappointed. We will reopen a factual error in a certificate, which is a different object. See Certificates.
Ready to bring
a team to Singapore?
Applications for the next TGA cohort open each September. Schools, learning centers, and independent teams of eight are welcome to apply.
Register a team See the Galaxy