IQ IqraLearning

PropTech

PropIntel NC | Investment Property Intelligence

Distressed-deal intel with cited AI proposals and rule-based math.

Atif Jaffery · AI Architect — cited tool-loop analyst and human-accept workflow

1 analyst agent · calculators own the numbers

01The situation

In 60 seconds

Business problem

North Carolina investors spend days stitching together courts, GIS, deeds, tax and Secretary of State data, miss auctions, and bid without the full math.

Unacceptable failure

A bid built on a number the model made up.

Design move

The analyst proposes cited inputs. The investor accepts or rejects each field. Rule-based calculators compute every number.

What it shows

Every proposed field links to its evidence, and the model never writes to the books.

The problem

Distressed-property deals move fast and the data is scattered across county systems. The goal is to analyze a deal in minutes instead of days, without trusting an uncited AI valuation.

Who feels it

  • Distressed-property investors (MVP: five NC counties)
  • Lenders and partners who receive the PDF report

What’s at stake

  • Missed auctions.
  • Bids without full flip or rental math.
  • Uncited AI valuations that can’t be trusted.

Constraints

  • Public NC sources; dev data is labeled synthetic until live pulls are approved.
  • The model can read, not write: a read-only tool loop.
  • Calculators are deterministic.

Failures this system must never produce

  1. F1

    A bid on an invented number

    No ARV, rent or cost figure enters the math without a citation and the investor’s acceptance.

  2. F2

    The model writing the books

    Accepted inputs feed the calculators; the model never overwrites them.

  3. F3

    Synthetic data passing as real

    Dev data is labeled synthetic, and eval labels must not pollute seeded rows.

02The approach

What was built & where AI assists

Each failure gets one mechanism, and each mechanism has proof further down this page.

Failure Mechanism The model’s role Proof
F1
Mechanism

The Deal Analyst runs a read-only tool loop and returns proposed inputs, each with citations into an evidence ledger.

The model’s role

Proposes and cites. It can’t apply anything.

Explorer →
F2
Mechanism

The investor accepts or rejects each field. Calculators re-run on accepted inputs only.

The model’s role

None. The math is rule-based.

Guardrails →
F3
Mechanism

Rows carry data_origin = synthetic until live pulls are approved.

The model’s role

None.

Guardrails →

Where the model works, and where it doesn’t

Model-assisted

  • AI Deal AnalystLocal model by default, or Claude, OpenAI, Gemini or DeepSeek. Read-only tools.
  • NarrativeMemo grounded in calculator output and cited evidence.

Rule-bound or human

  • Data pipelineNormalizes and matches public county sources.
  • CalculatorsFlip, rental, Cash Bid and Invest math.
  • Field acceptanceThe investor accepts or rejects each proposed input.
  • PDF reportLender-ready property intelligence packet.

AI touchpoints and who decides

One analyst agent proposes. The investor decides field by field, and calculators own every number.

Click any card for its contract: inputs, outputs, decision authority, and guardrails.

LLM model-assisted Code deterministic Human person in the loop
Collects

Data

Public county sources, normalized and matched.

Proposes

Analysis

Reads through a read-only tool loop and cites everything.

Decides

Authority

The investor decides each field, then the bid.

Computes

Math

Deterministic numbers and a lender-ready report.

Guardrails & trust

  • Cited proposals only; human decision step before inputs apply.
  • Synthetic data labeled in build environments.
  • Model does not overwrite calculator authority without acceptance.

03See it work

See it run

A 30-second walkthrough of one distressed property, and a live demo on synthetic county data.

Live demo · cloud

Coming soon

Underwrite a deal without trusting a guess

  • Open a sample distressed property
  • Ask the analyst for proposed inputs
  • Check each citation in the evidence ledger
  • Accept some fields and reject others
  • See the Cash Bid and rental math re-run
Open live demo · soon Book a walkthrough
  1. 0:00–0:06 Property

    A filing turns into a property record with county data attached.

  2. 0:06–0:14 Propose

    The analyst proposes inputs, each with a citation.

  3. 0:14–0:22 Accept

    The investor accepts some fields and rejects one.

  4. 0:22–0:30 Compute

    Calculators re-run Cash Bid and rental math.

Interactive architecture

Walk a real request through agents, services, stores, and human gates. Click any node for its contract.

04The result

What changes for the investor

Faster path from filing to underwritable view

County data stitched together for you.

Numbers you can audit

Calculators own the math; the model only proposes (F2).

A clean handoff to lenders

A PDF report with cited inputs.

How I check it behaves

Each check has a definition and a target. Results appear only from real runs, with the set size and date.

Check What it counts Target Last run
Uncited proposals F1 Proposed fields without a citation in the evidence ledger 0 Run pending
Model writes F2 Calculator inputs changed without the investor’s acceptance 0 Run pending
Label integrity F3 Synthetic rows shown without the synthetic label 0 Run pending
Real-deal suite F1 Proposals checked against the real-deal evaluation suite Tracked Run pending

05Looking back

Architecture decisions

  • Separate proposal from calculation F2

    So the numbers stay deterministic.

    Cost acceptedThe investor has to review every field.

  • A focused NC MVP before multi-state expansion

    So the data pipeline is solid where it runs.

    Cost acceptedA smaller market at launch.

Known limits and next step

Scope · disclosed

The MVP covers five NC counties, and real inventory is gated: development runs on synthetic data.

Expansion beyond the MVP counties is designed but not built.

Next step

  1. Approve live pulls county by county.
  2. Expand toward all 100 NC counties, then other states (designed, Phase 2–3).

Project demo