Aandar — Real-estate finance advisory
AI-assisted advisory platform for real-estate finance: chat, transactions, and team — all CMS-driven.

About
Aandar is a real-estate finance advisory platform with an AI chatbot that answers domain-specific questions from a curated, domain-tuned system prompt — alongside live commercial-real-estate interest-rate widgets, a client-facing loan sizer, transactions, team profiles, and lead capture, all Sanity-driven. The AI layer is provider-agnostic: it calls Claude, Gemini, OpenAI, or Groq over raw HTTP and is swappable with a single env var, so the team is never locked to one vendor's SDK, pricing, or SLA.
Problem
A boutique advisory needed a credible web presence with real-time transaction data, an answerable chatbot for prospects, client-facing financing tools, and a private area the team could run themselves — on a CMS they could actually use.
Approach
Next.js App Router for the public site, Sanity for editorial content and structured data, and a thin AI orchestration layer that routes prompts to whichever provider is best (and cheapest) at the time. Financing tools pull live Fed rate data and CMS-managed market figures; a lightweight custom auth (bcrypt + JWT) gates the team dashboard.
Outcome
A site that doubles as a sales surface, a content platform, and a client tool — fast, editor-friendly, and resilient to AI-provider price/SLA changes.
What's interesting
- AI chatbot answering CRE-finance questions from a domain-tuned system-prompt knowledge base — no vector DB needed
- Provider-agnostic AI — swap Claude / Gemini / OpenAI / Groq with one env var, each called over raw HTTP (no vendor SDK lock-in)
- Live interest-rate widgets — SOFR, Prime & Treasuries via the Federal Reserve FRED API, hourly-cached
- Client-facing loan/deal sizer driven by CMS-managed market data
- Authenticated team dashboard — bcrypt password hashing + JWT session cookies, users stored in Sanity
- Sanity-driven content (team, transactions, posts, leads) with on-demand ISR revalidation via tagged fetches
- Rate-limited public APIs; AI keys stay server-side only