BUILD WITH AI 2026: THE FABS 3D AI CHALLENGE

Hosted by GDG on Campus KIMEP × FABS3D AI

A high-stakes, asynchronous DeepTech + Business competition at the frontier of Spatial AI. 3 tracks · 12 days · $1,000 prize pool.

Detailed tasks information can be found here:
https://fabstoryai.com/gdgapril

─────────────────────────────────

🗓 TIMELINE

─────────────────────────────────

Build Phase (Remote) Apr 21 – May 2

Submission Deadline May 2, 23:59

Pitch Day @ KIMEP May 3 · 12:00–18:00

No live coding on Pitch Day. Bring a working demo + polished deck.

─────────────────────────────────

☁️ GOOGLE CLOUD CREDITS

─────────────────────────────────

$5 Google Cloud Credits for all participants (Vertex AI + Gemini API). RSVP with your Gmail before Apr 27 to claim.

─────────────────────────────────

Use this Google form link for submissions: https://forms.gle/dcSy8yz23pnwCjyWA

─────────────────────────────────

TRACK 1 — 3D CAD Pipeline · $300

─────────────────────────────────

Build a CAD app that ingests a dirty 3D model (CBCT, STL, OBJ), auto-cleans the mesh, shows a 3D preview, and exports a clean STL. Think Photoshop Auto-Enhance for spatial data. Any domain welcome — dental, prosthetics, architecture, jewelry.

Reference files (face scan + CBCT):

drive.google.com/drive/folders/1c_yS-N6obJZ5Cf0KMUBjp8rssc4Oljjq

REQUIRED

• File upload (drag-and-drop, multi-format)

• Auto-cleaning (noise, artifacts)

• 3D interactive preview

• Export to STL

BONUS

★★★ Auto-Layering / AI Mesh Generation / Auto-Alignment

★★ Dashboard (vertices before/after)

★ MCP integration

STACK: VTK · Open3D · trimesh · Three.js · Blender API

JUDGING: end-to-end pipeline → cleaning quality → UI/UX → README

CHECKLIST

[REQ] Working CAD app · Clean STL · Demo (video/GIF) · Source links

[BONUS] Auto-Layering · AI Mesh Gen · Auto-Alignment · Dashboard · MCP

🥇 $150 · 🥈 $100 · 🥉 $50

─────────────────────────────────

TRACK 2 — ML & Autonomous Agents · $300

─────────────────────────────────

Train an RL agent to navigate a 3D maze as fast as possible. Agent sees 23 inputs, controls a continuous force vector. Build everything yourself: maps, environment, benchmark, agent.

WORLD PHYSICS

🛣️ Asphalt ×1.0 · 🌿 Grass ×0.7 · 🏖️ Sand ×0.4 · 🧊 Ice ×1.2

❄️ Cold: energy ×1.5 · 🔥 Heat: speed ×0.6 · 📐 Terrain: slope (−1..+1)

AGENT

Observation (23): 8 distance rays + 8 surface rays + compass (2) + floor/slope/temp (3) + velocity (2)

Action: force_x, force_y (−1.0..+1.0) — continuous control, not discrete buttons

TRAIN/TEST SPLIT (required)

Min 5 maps: 3 train + 2 test. Agent never sees test maps during training. Judges may test on their own OOD map.

KEY CONCEPT: Reward shaping — default sparse reward (+10 finish, −0.1/step) won't train. Design your own intermediate rewards. Worth 10/35 design points.

SCORING (100 + bonuses)

⚡ Speed: ≤60 steps=40pts · 61–80=30 · 81–120=20 · 121–200=10 · DNF=0

🧠 Design (35): Sensing≤10 · Reward shaping≤10 · ML justification≤8 · Generalization≤7

📈 Presentation (25): Learning curve≤8 · Before/After≤7 · Live demo≤5 · README≤5

✨ Bonuses: +5 Heatmap · +5 Route physics · +5 3D viz · +3 FABS3D

CHECKLIST

[REQ] GitHub+README · Maps (train/test split+links) · Model .pt/.zip · Benchmark · Learning curve · Before/After · Approach description

[BONUS] Live demo +5 · Heatmap +5 · FABS3D +3

STACK: Python · PyTorch · Stable-Baselines3 · Gymnasium · NumPy · Google Colab

🥇 $150 · 🥈 $100 · 🥉 $50

─────────────────────────────────

TRACK 3 — Product & Business Strategy · $300

─────────────────────────────────

You are not coding. You are selling. Take the spatial tech from Track 1 or 2 and prove it's a business. Built for Data Analytics, Finance, and Business students — no PyTorch required.

THREE TASKS

🟢 EASY — ICP: Who buys? What's their pain? How much do they pay?

Best method: interview 10 people who already paid to solve this problem.

🟠 MEDIUM — Competitors: Who else solves this? Where do they fail?

Best method: interview real customers of your direct competitors.

🔴 HARD — Market commitment (best → worst):

Pre-order > product testing > 30min offline interview > 15min online interview > waitlist

More and higher quality = more points.

⚠️ Pretty slides without data = 0 CustDev points. Judges are founders and investors — they spot GPT insights instantly.

SCORING (100 + bonuses)

🤝 CustDev (40): Commitment quality≤20 · ICP precision≤12 · Competitor analysis≤8

📊 Strategy (35): GTM≤15 · Unit Economics≤12 · TAM/SAM/SOM≤8

🎤 Presentation (25): Pitch≤10 · Evidence≤8 · Deck+Doc≤7

✨ Bonuses: +8 Real sales/pre-order · +5 Financial Model · +5 LOI/Partnership · +3 Fabs3D use case

CHECKLIST

[REQ] ICP Document · Competitor Matrix · CustDev Evidence · GTM Strategy · Unit Economics + TAM/SAM/SOM · Pitch Deck (max 10 slides)

[BONUS] Real sales +8 · Financial Model +5 · LOI +5 · Fabs3D +3

TOOLS: Notion · Google Slides · Google Sheets · Typeform · Miro · G2/Capterra · LinkedIn

🥇 $150 · 🥈 $100 · 🥉 $50

─────────────────────────────────

🏆 BEST OVERALL TEAM — $100

─────────────────────────────────

Open to any track. Single strongest submission combining technical depth + business impact. Cross-track teams strongly encouraged.

TOTAL POOL: T1 $300 + T2 $300 + T3 $300 + Best Overall $100 = $1,000

─────────────────────────────────

Full track details: fabstoryai.com/gdgapril
Use this Google form link for submissions: https://forms.gle/dcSy8yz23pnwCjyWA

GDG on Campus KIMEP × FABS3D AI · Almaty · May 3, 2026

Registration deadline: April 27 (got extended)