Tanseeq / تنسيق

AI Development Conditions Briefing Officer · Track: Decision Intelligence

Tanseeq (Arabic: تنسيق) means coordination, arrangement, organizing.

Tanseeq reviews one vacant parcel and one proposed real estate development, then produces an advisory conditions brief for a human review committee. It is not a regulatory approval tool, not a parcel-scoring app, not a generic chatbot, and not only a zoning checker. It is a Decision Intelligence prototype that coordinates land, capital, market, community and mobility signals into one review brief.

The committee sees one of four advisory positions:

Decision readiness	Meaning
Support for review	Signals are aligned; can be supported substantially as submitted.
Support with light conditions	Minor coordination gap; a couple of light conditions.
Support with conditions	Material gap between development pressure and community absorption; conditions required.
Re-scope before review	The gap is wide enough that the scheme should be revised before committee.
Hold for additional evidence	Inputs are incomplete; Tanseeq will not take a position.

---

The problem

Abu Dhabi is growing fast, and every new development adds pressure on the surrounding community. A parcel can look ready because the land is vacant, the market looks active and investors are interested — but if nearby services, amenities and mobility access are not ready, the project may create future pressure for residents and operators.

Today these signals are usually reviewed separately: land potential, investor fit, market momentum, service demand, amenity supply, mobility readiness. Tanseeq connects them into one coordinated review so a human committee can see whether a proposal should be supported as submitted, supported with conditions, re-scoped, or held for more evidence.

The solution

A deterministic, auditable reasoning core runs five coordination checks, computes development pressure vs community absorption, derives the advisory decision label, and assembles a grounded evidence packet. An optional AI layer then writes the committee-style narrative using only that evidence packet — never inventing facts and never claiming approval. With no API keys, a complete deterministic brief is produced instead, so the app never breaks.

The five coordination checks

1. Land / Value — vacancy, use alignment, development potential, infrastructure, value vs district transaction comps.
2. Demand / Supply — service demand index, resident experience, OSM amenity counts (healthcare, education, retail, community, services, parks), added service pressure.
3. Mobility / Access — community mobility score, OSM transit amenities (bus stops/stations, fuel, parking), residential density, added access pressure.
4. Capital Fit — investor mandates matching district + sector, capital range vs parcel value, risk profile, horizon, strategic fit. Surfaces the matching investors.
5. Market Momentum — 2023–2026 transaction trend and price/sqm, listing-status absorption, district yield.

Core scoring model

development_pressure   = land_signal*0.28 + capital_fit*0.22 + market_momentum*0.22
                       + infrastructure_readiness*0.18 + proposal_intensity*0.10

community_absorption   = amenity_support*0.25 + mobility_access*0.25 + resident_experience*0.20
                       + inverse_service_demand*0.20 + community_facility_support*0.10

coordination_divergence = development_pressure − community_absorption

Divergence bands → decision: 0–15 Support for review · 16–30 Support with light conditions · 31–50 Support with conditions · 51+ Re-scope before review. Missing/incomplete inputs → Hold for additional evidence. Decision priority (ROI / community / balanced) tilts the thresholds.

Demo flow

1. App auto-selects a strong demo case: a vacant parcel with high development potential, strong land/capital/market signals, but weak community absorption and constrained mobility → high coordination divergence. (Live demo lands on PRC-0096, Al Ghadeer: potential 87, capital fit 95, market 65 — but demand 36, mobility 26 → pressure 69 vs absorption 34.)
2. Choose proposed use and scenario settings (units, retail share, community facility, mobility condition, decision priority).
3. Run Tanseeq Review → five checks, divergence meter, conditions brief.
4. Scenario Lab — compare A (as submitted) · B (add facility) · C (add mobility) · D (both) · E (phase units + conditions). Divergence and decision update live.
5. Download / print the advisory conditions brief.

Data used

All datasets live in /data so the app runs and deploys independently (no absolute paths at runtime):

File	Used by
districts.csv	Land/Value, Market
sample_parcels.csv	Land/Value, Capital, demo selection
sample_transactions.csv	Land/Value (comps), Market (trend)
sample_investors.csv	Capital Fit
sample_communities.csv	Demand/Supply, Mobility, absorption
sample_listings.csv	Mobility (density), Market (absorption)
osm_amenities.csv	Demand/Supply, Mobility

Challenge sample datasets are synthetic and used for prototype scoring. osm_amenities.csv is real OpenStreetMap data © OpenStreetMap contributors (ODbL) — attribution shown in the UI and exported brief.

How AI is used

• The deterministic layer computes the five checks, pressure, absorption, divergence, decision label and the evidence packet.
• The AI layer generates the committee-style explanation (summary, why-not-as-submitted, required conditions, evidence references, questions for human review, limitations) from only the evidence packet, with a strict safety rule: use only the evidence, invent no statistics, claim no legal/regulatory approval, say "insufficient evidence" when data is missing, stay advisory.
• The decision label is always pinned to the deterministic result — the model can rephrase but never override it.

What works without API keys

Everything. With no keys the app produces a complete, grounded deterministic conditions brief. The AI layer is strictly additive.

Optional LLM environment variables

Variable	Effect
ANTHROPIC_API_KEY	Use the Anthropic API (claude-opus-4-8 by default).
ANTHROPIC_MODEL	Override the Anthropic model id.
OPENAI_API_KEY	Used if no Anthropic key (gpt-4o-mini by default).
OPENAI_MODEL	Override the OpenAI model id.

Provider order: Anthropic → OpenAI → deterministic fallback. Any provider error falls back silently.

How to run

npm install
npm run dev      # http://localhost:3000

Command	Description
npm run dev	Development server.
npm run build	Production build (type-checks the whole project).
npm run start	Serve the production build.
npm run lint	ESLint.
npm run test	Vitest suite (CSV parsing, scoring, decision labels, demo selection, scenario divergence, deterministic brief).

Tech stack

• Next.js 16 (App Router) + React 19 + TypeScript
• Tailwind CSS v4
• Vitest for the reasoning-core tests
• Server routes POST /api/review (full review + brief) and POST /api/scenarios (Scenario Lab comparison)

Project structure

data/                         # CSV datasets (copied in, app-local)
src/
  app/
    page.tsx                  # Portal shell + intelligence room composition
    api/review/route.ts       # Full review (engine + optional LLM)
    api/scenarios/route.ts    # Scenario Lab comparison (engine only)
  components/
    CaseIntake.tsx            # Government-service intake form
    DecisionRoom.tsx          # Dark Cursor-style dashboard
    CheckCard.tsx             # One of the five coordination checks
    DivergenceMeter.tsx       # Pressure vs absorption + decision
    ScenarioLab.tsx           # A–E scenario comparison
    ConditionsBrief.tsx       # Government-style conditions brief
    ui.ts                     # Tone / decision styling
  lib/tanseeq/
    types.ts                  # Domain model
    csv.ts                    # Dependency-free CSV parser
    data.ts                   # Dataset loader + demo-case selection
    scoring.ts                # Five checks → pressure → divergence → label
    evidence.ts               # Evidence packet + deterministic brief
    llm.ts                    # Optional Anthropic / OpenAI layer
    scenarios.ts              # Scenario Lab presets
    export.ts                 # Markdown brief renderer
    tanseeq.test.ts           # Unit tests

Design

A fusion of TAMM Abu Dhabi government services (off-white official portal shell), a Cursor-style dark intelligence command center (the Decision Room), and premium real-estate intelligence — teal accents, subtle gold, Arabic/English microcopy, rounded service cards, status pills, evidence chips, animated meters.

What was built during the hackathon

• A new, fully Decision-Intelligence reasoning core (the previous build was a deterministic zoning compliance officer — that concept and all "approval-ready / Proceed / Approve with Conditions" wording was removed).
• Five coordination checks over six real challenge datasets + real OSM amenities.
• The pressure-vs-absorption-vs-divergence model and the advisory decision labels.
• Deterministic, grounded conditions brief + optional Anthropic/OpenAI narrative layer.
• Government-portal + dark-command-center UI, Scenario Lab, and Markdown/PDF export.
• A Vitest suite covering parsing, scoring, decision mapping, demo selection, scenario divergence and the no-key fallback brief.

Cursor usage summary

The initial scaffold was generated in Cursor but built the wrong concept (a zoning-compliance officer). This iteration refactored it into the final Decision Intelligence brief: replacing the engine, data layer, API, components and tests; wiring the real challenge CSVs into /data; and adding the optional LLM layer with a deterministic fallback.

---

Tanseeq's outputs are advisory and support — but do not replace — a human review committee. They do not grant statutory consent or approval.