DealLens combines public-record data, MLS feeds, and an LLM-powered comp engine to generate institutional-quality valuations. Acquisition analysts at small and mid-cap REITs use it to triage 50 to 200 deals a week down to the 3 worth a real underwriting session.

Acquisition teams were drowning in deal flow but bottlenecked on analyst hours. Existing tools (CoStar, Reonomy) gave them data but not synthesis. They wanted "give me a 1-pager I can actually act on", and they wanted it grounded in defensible comps.

An AI comp engine that selects the closest 8 to 12 comps from public + MLS data and explains why they were chosen (radius, vintage, condition, sale recency).

A 1-page deal report: valuation range, suggested cap rate, top 3 risks, and a markdown rationale.

Pipeline mode: drop in a CSV of 50 deals, get a ranked dashboard with quick-decline / quick-promote actions.

Slack + email notifications for deal owners with the AI rationale embedded.

Frontend: Next.js + TanStack Query + Recharts for analyst-friendly dashboards

Backend: FastAPI + Postgres + pgvector + Redis

AI layer: GPT-4 class model for narrative + a fine-tuned classifier for comp eligibility

Data: ATTOM, public records, custom MLS scraper layer with provenance tracking

Infra: AWS (Fargate + RDS + S3), Datadog for observability

Analyst capacity up from 18 deals/week to 70 deals/week.

Deals that made it to full underwriting closed at 2.4x prior rate (better top-of-funnel quality).

AI valuations within 4.2% of internal expert valuations on a 200-deal test set.