RA
All Demos Demo 1 of 23
← PrevNext →
AI Architecture
8 AI Systems. 3 Models.
One Intelligence Layer.

The only PropTech platform with cross-session semantic memory, AI semantic caching, multi-model cost routing, and a proprietary SLM pathway. Nelo gets smarter AND cheaper with every conversation.

The AI Control Tower — 8 Interconnected Systems

No competitor has more than 2 AI systems. RealRiches has 8 — each purpose-built, interconnected, and reinforcing. Together they form the most comprehensive AI infrastructure in PropTech.

Concierge

1. Nelo AI

Claude Sonnet 4.6 with native tool use, pgvector 1536-dim memory, 4 memory types. Cross-session recall via embedding similarity. 6-portal intelligence — sees tenant, landlord, agent, vendor, HOA, and admin context simultaneously. No competitor has cross-portal AI.

Voice

2. Voice Pipeline

Twilio receives call, streams to Deepgram nova-3 STT, transcript feeds Claude for reasoning, response streams to ElevenLabs eleven_flash_v2_5 TTS. Sub-300ms target. 10 NYC-priority languages with dedicated voice IDs. 6,673 lines of production code across 5 packages.

Orchestration

3. LangGraph Orchestrator

8 specialized agents with PostgresSaver checkpoint persistence. Full lead-to-lease pipeline: qualify, schedule tour, review application, analyze credit, draft lease, calculate deposit, coordinate move-in, check compliance. Each agent has defined tools and model routing. 3,794 LoC.

Verification

4. Z3 Policy Gate

Formal verification across 6 policy domains using Microsoft Z3 satisfiability solver. Mathematically proves business rules do not create protected-class disparate impact BEFORE execution. Not statistical sampling. Mathematical proof.

Screening

5. FairScreen ISP

3-model consensus Independent Screening Pipeline. A screening decision only proceeds when all 3 models reach agreement. Any disagreement routes to human review. Full audit trail with disparate impact analysis per decision. 8,529 lines.

Cryptography

6. ZKP (Groth16)

Zero-knowledge proof tenant qualification. Tenants prove they meet income and credit thresholds without revealing actual financial data. Landlord learns only yes/no. Eliminates the data surface for discrimination. Patent-eligible with confirmed zero prior art.

Identity

7. DAV (ISO 18013-5)

Digital Age Verification via mobile driver license standard. Government-grade identity verification without exposing unnecessary personal data. Fraud prevention built into every transaction flow.

Content

8. AI Copilot

Listing descriptions, property flyers, social media posts, email campaigns, open house materials — generated in seconds. Tri-LLM routed: cheap-first for drafts, premium escalation for polished output. Subscription-tier aware.

Tri-LLM Routing Architecture — 18,360 LoC Verified

Every competitor uses one AI model for everything. RealRiches routes by capability, cost, and subscription tier across three providers — with automatic fallback, health monitoring, and circuit breaking.

0.8x CHEAPEST
Gemini
Gemini 2.5 Pro
Google · Research & Ops (Gemini 2.5 Pro)
• Lead enrichment & Deep Research artifacts
• Spreadsheet ops & browser automation
• Multilingual intake (10 languages)
• Campaign surfaces & market intelligence
• 13 research artifact types
Default for all research & ops tasks
1.0x PRIMARY
Claude
Claude Sonnet 4.6
Anthropic · Premium Synthesis (Sonnet 4.6)
• Nelo AI core intelligence (Opus 4.6)
• Investor memos & compliance analysis
• Long-form narratives & lease drafting
• Polished outbound & seller reports
• Safety-critical decisions & reasoning
Nelo's brain — premium escalation tier
1.5x FALLBACK
OpenAI
GPT-5.4
OpenAI · Copilot & Orchestration (GPT-5.4)
• Operator assistant & workflow orchestration
• General tasks & mixed tool-using flows
• File-assisted tasks & code generation
• Automatic failover from Claude/Gemini
• Circuit breaker fallback provider
Redundancy — zero single-provider dependency

The SLM Flywheel — AI That Gets Cheaper With Scale

Nelo is simultaneously the production AI concierge today AND the data collection pipeline for tomorrow's proprietary model. Every conversation, every correction, every cache hit feeds a flywheel that reduces costs while improving quality.

StageExternal LLMCost/QueryGross MarginMechanism
Year 0 (Now)100%$0.00380.7%Full external API (Claude + Gemini + ChatGPT)
Year 130%$0.000888%Semantic cache handles 70% of queries
Year 210%$0.000393%Fine-tuned SLM handles remaining simple tasks
Year 3+5%$0.000195%+Proprietary PropTech SLM trained on millions of interactions
Layer 1

Continuum Memory (459 LoC)

pgvector 1536-dim embeddings store every interaction. 4 memory types: user preferences, property interests, communication style, search patterns, behavior patterns, domain knowledge, feedback — plus 7 fast memory types for session context. Nelo remembers you 3 months later.

Layer 2

Semantic Cache (628 LoC)

pgvector-backed query/response cache with cosine similarity matching. When tenant B asks the same question tenant A asked, the cached answer returns instantly — no LLM call. 70% hit rate at maturity. Each new user warms the cache for future users. Network effect.

Layer 3

Cost Routing (3,388 LoC)

Every task type maps to cheap, standard, or premium candidates. Simple FAQ routes to GPT-5.4-mini ($3/MTok input). Compliance analysis routes to Claude Sonnet 4.6 ($3/MTok input). Budget ledger tracks per-model costs. Token governor enforces input/output limits per quality tier.

Layer 4

Subscription Entitlements (281 LoC)

5 subscription tiers (free through enterprise) with per-feature AI budgets, premium escalation quotas, and feature-level overrides. The $89 tier gets quality via cheap-first routing. The $149 tier gets richer premium narratives. Enterprise gets unlimited.

Competitor AI Comparison

AI CapabilityRealRichesEliseAI ($2.2B)RealPageYardiAppFolio
Cross-Session Memory4 types, pgvectorSession-onlyNoneNoneNone
Semantic Cache70% hit rate at maturityNoneNoneNoneNone
Multi-Model Routing3 models + cost trackingSingle modelSingle modelNoneRealm-X AI Copilot
AI Tier Monetization5 tiers (FREE to ENTERPRISE)One tierNoneNoneNone
Formal Verification (Z3)6 policy domainsNoneNoneNoneNone
Zero-Knowledge ProofsGroth16NoneNoneNoneNone
Voice AI PipelineDeepgram + Claude + ElevenLabsProprietary NLPNoneNoneNone
Agentic OrchestrationLangGraph 8-agentIntent classifiersNoneNoneNone
Multi-Language Voice10 languages + dedicated voices33 spokenNoneNoneNone
SLM Training Pathway51 Prisma models collecting dataLocked to proprietary NLPNoneNoneNone
LLM Price Resilience85% absorbed via cache100% pass-through100%N/AN/A
AI Systems Count8 systems2 systems1 system01 system

The Investor Line

"Every competitor using raw API calls is permanently locked at $0.003/query. RA's cost drops 97% over 3 years while quality improves. If LLM prices rise 50%, RA absorbs 85% of the impact. Competitors absorb 0%. RA's gross margin trajectory: 80.7% to 88% to 93% over 3 years with zero pricing changes."

Platform and Partners FARE Act Compliance
RAHome