The Platform

The Only PropTech OS Where Compliance Violations Are Architecturally Impossible.

1.92 million lines of TypeScript. 419 Prisma models. 8 production AI systems. 22 market toggles. FARE Act, LL97, LL144, Colorado ADAI — not bolt-on modules but structural enforcement at the API layer. Built to deploy, maintain, upgrade, and scale faster than anything in PropTech.

1.92M

Lines of Code

321

Packages

AI Systems

2,312

API Endpoints

419

Data Models

16/16

CI Green

Section 01 — Platform Architecture

Every Market and Every Feature Is Independently Launchable

RA's verticalGate() middleware enforces 3 layers on every API call: MarketConfig → FeatureFlag → Entitlement. A university buys ONLY CARA compliance. An HOA buys ONLY elections. An agent subscribes to ONLY voice tools. No feature requires the full suite. 22 master toggles × 49 sub-feature keys = 71 switches per market.

Architecture Layer	Implementation	What It Controls
MarketConfig	Prisma model + features JSON	22 master markets: Luxury Rental, Student Housing, Senior Living, Campus, CRE Office, Industrial, Co-Living, STR, Military, Affordable, Mixed-Use, HOA, FSBO, Agent, PM, and 7 more
FeatureFlags + Redis	132-line FeatureToggleEngine	49 sub-features: FAIRSCREEN, CHARGEPOINT, ESUSU, STARSHIP, CAMPUS_ROBOTICS, VOICE_AI, 3DGS_TOURS, ALEXA, ZKP_SCREENING, + 40 more (CRE alone has 8)
Entitlement SKUs	22 EntitlementSKUs + billing	Subscription-tier access control. Agent Pro gets voice + copilot. Enterprise gets all 49. Granular billing per feature.
verticalGate() Middleware	On every API route handler	3-layer enforcement per request: is market enabled → is feature toggled → does subscription include this SKU

WhiteLabel PaaS Deployment WhiteLabelAdapter + percentage rollouts + dependency graphs + override hierarchies
Partner A (Zillow): toggles ON → HAL, FARE Act, 3DGS, Voice AI, Leads — toggles OFF → Campus, Senior, HOA
Partner B (University): toggles ON → CARA, SafeWalk, Robotics, HAL — toggles OFF → Luxury, Commercial, Agent
Partner C (Senior Living): toggles ON → HIPAA, Wearables, HAL, Voice — toggles OFF → Campus, FSBO, Agent

Section 02 — AI Control Tower

8 Production AI Systems. No Competitor Has More Than 2.

These are not roadmap items. They are production systems with real API wiring, Prisma persistence, and behavioral tests. The AI Control Tower orchestrates all 8 systems through a unified governance layer with cost tracking, model routing, and compliance audit trails.

Nelo AI Concierge

Tri-LLM routing (Claude + Gemini + GPT). pgvector continuum memory. 5 subscription tiers (FREE → ENTERPRISE). 13 Prisma models. Sub-300ms voice.

Voice AI Pipeline

Twilio → Deepgram STT → Claude reasoning → ElevenLabs TTS. Real-time transcription, sentiment analysis, automated CRM actions. Replaces $60K/yr leasing labor.

FairScreen ISP

3-model consensus screening. Disparate impact monitoring. FARE Act + Colorado ADAI + 17-jurisdiction compliance. SHA-256 audit trail. Patent pending.

ZKP Verification

Groth16 zero-knowledge proofs. Tenant proves income qualification without revealing salary. First in PropTech. Mathematically impossible to forge.

3DGS Virtual Tours

Phone-captured Gaussian Splatting. No $5,400 Matterport camera. iPhone → 4.2M splats → interactive 3D tour in 16 minutes. $0 hardware cost.

HAL Access Control

4-vendor abstraction: ButterflyMX, Brivo, Allegion, Seam. One API controls all doors. Self-guided tours, Alexa narration, emergency lockdown.

Z3 Formal Verification

Microsoft Z3 solver engine. Mathematical proof — not testing — that policies are compliant. 6 verification domains. Output: VALID/INVALID, not "probably."

LangGraph Orchestrator

8-agent autonomous workflow with PostgresSaver state. Lead → screen → tour → sign with zero manual steps. @langchain/langgraph 1.2.0.

Section 03 — Nelo AI: The Learning Flywheel

Every Conversation Makes the Platform Smarter AND Cheaper

Nelo is not a chatbot. It is a 3-layer learning system with persistent semantic memory, building-level cache intelligence, and a self-hosted SLM trajectory that drives marginal inference cost to $0. 7,515 lines across 8 packages. 13 dedicated Prisma models.

Layer	Technology	What It Learns	Cost Impact
Layer 1: Continuum Memory	pgvector (1,536-dim) + OpenAI embeddings	Per-user preferences, communication style, search patterns across months. 7 memory types (Fast + Global). Nelo never asks the same question twice.	30–50% token reduction
Layer 2: Semantic Cache	pgvector cosine distance (≥0.92 threshold)	Per-building query patterns. A 500-unit building reaches 60%+ cache hit rate in 90 days. At maturity: 70% of queries never touch an LLM.	60–70% queries eliminated
Layer 3: SLM Routing	Self-hosted model endpoint + fallback chain	High-frequency intents (rent status, maintenance, amenity hours) route to $0-cost SLM. Confidence <0.7 falls back to Claude transparently.	$0 marginal for 40%+ queries

The Nelo SLM trajectory is fully engineered — training data export (JSONL), PII scrubbing pipeline, model router with SLM slot, client adapter for vLLM/Ollama/ONNX. Deploying the fine-tuned model is a configuration change, not an engineering project.

Section 04 — TokeniMax: 3-Layer Cost Reduction Stack

$0.02/Query → $0.0031/Query. 84.5% COGS Reduction.

RA integrates its proprietary TokeniMax compression engine alongside semantic caching and intelligent routing. Three independent cost reduction vectors that compound — the published 80.7% gross margin is conservative.

AppFolio Gross Margin (FY2025 10-K)63.7%

63.7%

RA Platform Year 180.7%

80.7%

RA Platform Year 3 (with TokeniMax + SLM)90.1%

90.1%

COGS Formula — 3-Layer Stack COGS = (%miss × Cost_LLM × (1 − compression%) × routing_discount) + (%hit × Cost_cache)

At 60% cache hit, 40% TokeniMax compression, 37% routing savings:
= (0.40 × $0.02 × 0.60 × 0.63) + (0.60 × $0.00011)
= $0.00302 + $0.000066 = $0.0031/query (84.5% reduction from $0.02 baseline)

What this means to operating costs: Every AI feature your competitors add degrades their margins (linear LLM cost scaling). RA's margins improve with scale because cache hit rates climb, SLM coverage expands, and TokeniMax compression compounds. By Y3, RA processes the same query volume at 1/6th the AI cost of any incumbent.

Section 05 — Compliance Built Into DNA

Not Bolt-On Modules. Structural Enforcement.

RA doesn't "support" FARE Act compliance — it makes non-compliance architecturally impossible. Every screening decision passes through FairScreen ISP (3-model consensus) AND Z3 formal verification (mathematical proof). Every listing runs through disparate impact monitoring. No human can bypass the compliance layer because it's enforced at the API middleware level, not the UI.

Regulation	RA Implementation	Enforcement Level	Competitor Status
FARE Act (Federal)	FairScreen ISP: 3-model consensus + disparate impact tracking + SHA-256 audit	API middleware — cannot bypass	No competitor has automated this
Colorado ADAI	Algorithmic decision audit trail + bias testing + automatic reporting	Pre-decision gate — blocks non-compliant actions	Zero compliance automation
NYC LL97 (Carbon)	Energy module: real-time carbon tracking, penalty calculation, ESG reporting	Continuous monitoring	Manual spreadsheets
NYC LL144 (AI Audit)	Z3 formal verification: 6 domains, mathematical proof of policy compliance	Mathematical proof, not testing	No automated audit system
Fair Housing (17 jurisdictions)	Jurisdiction-aware screening rules, protected class monitoring, automatic flagging	Jurisdiction detection per property	Generic disclaimers only

93.3% of operators experienced application fraud in 2024 (NMHC Pulse Survey). 84.3% saw fabricated pay stubs. $4.2M average portfolio bad debt. * RA's ZKP verification makes document forgery mathematically impossible — the tenant proves qualification through zero-knowledge proof without revealing the underlying data.

Section 06 — Self-Evolving Architecture

SEAUS: The Platform That Upgrades Itself

627 lines of production TypeScript. SEAUS (Self-Evolving Autonomous Upgrade System) continuously scans, evaluates, and recommends upgrades across the entire 321-package monorepo. Zero human intervention for dependency management.

SEAUSScanner

pnpm outdated + pnpm audit across entire workspace. Risk classification (low/medium/high/critical) by semver. CVE detection with CVSS scoring and CWE classification. SHA-256 checksums seal every scan result.

UpgradeReporter

Actionable upgrade recommendations with effort estimates (trivial/small/medium/large). Automatic compatibility checking across 321 packages. Dependency graph awareness prevents breaking changes. Nightly automated scans.

Why this matters for partners: Yardi's 42-year codebase takes months to update a single dependency. RA's SEAUS identifies, risk-scores, and recommends every upgrade across 321 packages automatically. Better models produce better SLM training data. Each upgrade accelerates the margin flywheel from 80.7% toward 95%+.

Section 07 — Real API Integrations

16 Production Partners. Pre-Wired. Zero Left to Build.

Not "planned integrations." These are native API adapters with OAuth 2.0, webhooks, and Prisma persistence. A competitor starting from scratch needs 12–18 months to replicate this ecosystem.

StripePayments + Billing

PlaidBank Verification

EsusuCredit Building

ChargePointEV Charging

LemonadeRenters Insurance

RhinoDeposit Alternative

Sure / AssurantInsurance

TheGuarantorsLease Guaranty

TwilioVoice / SMS

DeepgramSpeech-to-Text

ElevenLabsText-to-Speech

Anthropic ClaudePrimary LLM

Google GeminiResearch LLM

OpenAI GPTCopilot LLM

Amazon AlexaSelf-Tour Voice

Zillow / CoStarListing Syndication

Section 08 — HAL: Hardware Abstraction Layer

4 Access Control Vendors. One Unified API.

RA's HAL makes vendor switching a configuration change, not an architecture rewrite. If Brivo raises prices, shift to Allegion via Seam. If Allegion restricts API access, Brivo panels fill the gap. No vendor can hold the platform hostage.

20K+

ButterflyMX Buildings

Video intercom. 1.5M+ daily users. OAuth 2.0 REST API. 18+ PMS integrations.

100K+

Brivo Locations

Post-Eagle Eye merger. AI cloud security. $192M SECOM backing. 80 countries.

$4.07B

Allegion Revenue

Schlage, Zentra, Von Duprin. Apple/Google Wallet. Institutional dominance.

29+

Seam Device Brands

Universal API. Yale, Dormakaba, August, Assa Abloy. 30-day hardware swap.

Section 09 — Competitive Gap Analysis

The Revenue Your Competitors Will Capture — If They Move First

Each platform below dominates its lane. But none can build what RA offers without 36–48 months and $40M+ in engineering *. The question is not whether this gap gets filled — it's who fills it first and locks out everyone else for 18–24 months.

Zillow Group — $2.58B Revenue

240M+ MAU. SkyTour 3DGS expanding. ShowTime scheduling. Massive consumer reach. FTC antitrust suit active.

No FARE Act compliance. No FairScreen. No Voice AI pipeline. No HAL integration. Zillow can list — but can't screen, tour, unlock, or comply.

Yardi Voyager — ~$1.6B Revenue

42-year-old codebase (1984). 16M+ units. Deep PMS entrenchment. Chat IQ basic AI. G2 Score: 71 (lowest major PMS).

Monolithic legacy architecture. Adding AI means years of rewriting, not months. No real-time voice, no 3DGS, no ZKP, no multi-vendor HAL.

AppFolio — $951M Revenue

9.4M units. Realm-X AI Copilot (industry's best PMS AI). G2: 95. But gross margin: 63.7%. Every AI call degrades it.

Structurally cannot match RA's 90.1% margin. No FairScreen, no ZKP, no 3DGS, no HAL, no energy module. Mid-market focus — no luxury depth.

CoStar — $3.2B Revenue

Apartments.com. Matterport acquisition ($1.2B). Massive CRE data. Expanding into multifamily operations.

RA delivers superior 3D tours from a smartphone at $0 hardware. The $1.2B Matterport acquisition is now a depreciating liability.

Entrata — $4.3B Valuation

Full-stack PMS. ResidentPay. Open API ecosystem. Blackstone-backed. Not rated for Autonomous Task Execution on G2.

Widest AI gap relative to valuation. No voice pipeline, no FairScreen, no 3DGS, no HAL, no ZKP. The most exposed incumbent.

EliseAI — $2.25B Valuation

a16z Series E. Best conversational AI in PropTech. 125% lease conversion. Expanding into healthcare.

$2.25B for one of RA's eight AI systems. No screening, no 3DGS, no access control, no compliance engine, no energy module. Healthcare expansion dilutes RE focus.

Section 10 — Five-Layer Moat Architecture

$12–25M and 36–48 Months to Replicate. And You Still Won't Have the Data.

Each layer has different decay characteristics. The combination creates protection that strengthens rather than erodes over time.

Moat Layer	Time to Match	Decay Rate	Description
Technical Complexity	36–48 months	Slow	1.92M lines TypeScript, 419 Prisma models, 321 packages, 2,312 endpoints. Replication cost: $12–25M. *
Regulatory Compliance	24–36 months	Accelerating	FARE Act + LL97 + LL144 + Colorado ADAI + 17 jurisdictions. Each new regulation strengthens the moat. Cannot be engineered around.
Data Network Effects	Never	Compounding	Nelo learns from every conversation. Semantic cache warms per building. SLM trains on production data. The data moat grows daily.
Partner Ecosystem	18–24 months	Moderate	16 pre-wired integrations. 4 HAL vendors. Each partnership has bilateral API dependencies that take months to negotiate and build.
Compliance Certification	12–18 months	Slow	FairScreen ISP + Z3 verification + ZKP screening create certifiable compliance. Passing audit is binary — you either have it or you don't.

Section 11 — First-Mover Window

18–24 Month Structural Advantage. The Window Is Closing.

The first platform to integrate RA locks out competitors — not because the code is hard to copy, but because the data moat is impossible to replicate. By the time a competitor starts building, Nelo knows 3 million tenant interactions their AI has never seen.

18–24 mo

First-Mover Lead

Data compounds daily. Cache hit rates climb. SLM coverage expands. Every interaction makes the moat deeper.

$40M+

Second-Mover Cost

Engineering + data collection + compliance certification + partner ecosystem. 48–72 months from scratch.

Competitors with All 8

No platform — at any price or valuation — ships all 8 AI systems, toggle architecture, and compliance DNA today.

* All RA internal projections and revenue estimates are founder models based on bottom-up financial analysis. All data sources and methodologies listed at bottom of page. External data points sourced from: ButterflyMX 2025 Year in Review, AppFolio FY2025 10-K, Zillow FY2025 10-K, CoStar FY2025 10-K, Allegion FY2025 10-K, NMHC 2024 Pulse Survey, Seam.co. Entrata valuation from Blackstone public disclosure. EliseAI valuation from a16z announcement.