HeavyModel
AI as your core business system. A fully customized AI that becomes the engine of your investment workflow — replacing manual underwriting, research, and the analyst stack — and scaling the operation with you.
Ask a standard LLM the same deal ten times. You will get ten materially different answers.
This is the hidden defect of every general-purpose model — ChatGPT, Claude, Gemini, Perplexity included. Same prompt, same data, two minutes apart. The model re-samples the world. Cap rates drift. Comp pulls drift. Assumptions get re-imagined. Your underwriting is non-deterministic theatre.
Heavy doesn’t move.
The Deep Verification Layer.
Heavy is not a prompt. Heavy is a stack — a deterministic anchor wrapped around the language model, designed to remove the model’s ability to drift. Every number that leaves the system has been independently re-derived, cross-validated against primary sources, and signed with a confidence score.
What you see is not the model’s opinion. It’s the model’s opinion after being interrogated by four others, then reconciled against the data of record.
Comps as geography understands them. Not as a search engine returns them.
Every other model picks comps by ZIP code, neighborhood name, or a flat radius circle on the map. Heavy picks comps the way a local broker who has walked the streets would: by drive time, walk time, route friction, and — critically — by demand pocket. No general-purpose LLM does this. None of them know that a railway corridor between two assets makes them members of different rental universes.
“Asset B is 320 meters from Asset A — same neighborhood, same micro-market. Pull rent comp.” The model anchors your underwriting on a comp that is, in functional reality, in a different city.
320m line-of-sight — but no pedestrian crossing of the rail corridor for 3.2km. Real drive distance: 14 min, 7.8km around. Different demand pocket. Rent regime differs by 9 €/sqm. Comp rejected.
The data that won’t show up in any chatbot. We have it.
Standard LLMs are trapped behind Google. They can only see what was crawled, indexed, and cached — which excludes most of the data a serious investor relies on. Government registries, paid platforms, primary records, foreign-language portals: none of it lives in a general model’s training corpus or context window. It lives in ours.
The seller’s offering memorandum quoted average in-place rent at 11.40 €/sqm and 96.5% occupancy across an eight-building stabilised portfolio in the Plagwitz and Lindenau Stadtbezirke. A standard LLM, asked to validate, returned an “in line with Saxony rental averages” verdict, sourcing a Wikipedia summary and a 2022 industry report.
Heavy went elsewhere. It pulled the live Leipziger Mietspiegel 2024 at the Stadtbezirk level (not indexed by Google), cross-referenced eight Stadt Leipzig Bauordnungsamt permit records confirming building age and modernization status, and reconciled against twelve recent sublease filings from the Amtsgericht Leipzig — Grundbuchamt registry — all in German, all behind state portals.
Verified average: 10.85 €/sqm — not 11.40. True occupancy: 93.8%, not 96.5. Combined 7.4% NOI overstatement in the OM. On a €42M / $45M deal at a 4.4% gross yield, that’s ~$3.3M of value the seller was pricing into your equity. Heavy caught it. The OM became a negotiation document, not a thesis.
The sponsor’s package quoted $1,985 per unit average effective rent at 95% physical occupancy, with a “100% renovated interior” story across all 312 units. A standard LLM, given the same package, accepted the claim — its training on Charlotte rents was 18 months stale and its rent-comp coverage thin.
Heavy pulled fresh primary records: the Mecklenburg County Register of Deeds for actual loan history and recorded leasing covenants, Charlotte-Mecklenburg Planning for unit-upgrade permits (which showed only 38% of units pulled renovation permits since 2021, not the 100% the OM implied), HUD CHAS & Section 8 voucher payment standards for the census tract, and live institutional-grade effective-rent comps at the property’s exact tier and vintage.
Reality: $1,840 per unit effective rent, 91.4% true occupancy, and twelve units quietly carrying HUD vouchers below the claimed market rents. NOI overstated by 8.4%. At a 5.5% market cap rate, that’s ~$6.5M of overpriced value baked into the asking number. Bid revised to $70.5M with conditions; deal won at $71.2M.
A vacant infill parcel listed at $18.5M, with the seller’s marketing claiming “by-right entitlement for 340 units of multifamily.” A standard LLM accepted the unit count at face value and computed a residual land value that comfortably justified the price — at $54K per door, the math worked.
Heavy ran the parcel through the actual entitlement stack: the City of Austin Land Development Code (Compatibility Standards triggered by the adjacent SF-3 zoning, reducing usable building envelope by 22%), the Austin Watershed Protection impervious-cover cap for the Boggy Creek tributary basin, the Travis Central Appraisal District (CAD) for verified parcel boundaries and easement chain, and the city’s Heritage Tree Ordinance registry showing four protected oaks on-site requiring 30-foot critical root zones.
Real by-right yield: 268 units, not 340. Same RLV per door applied to the corrected count: $14.5M — a $4M (21.6%) overstatement on the land alone. The seller wasn’t lying; they were quoting a developer’s pro forma. Heavy quoted the code.
A regional bank was offered a $80.6M senior loan at 65% LTV against a 580,000 sqft single-tenant logistics facility valued by the sponsor at $124M, anchored by a 12-year NNN lease to a Fortune 500 logistics tenant. The credit memo was clean. A standard LLM-driven pre-screen returned green.
Heavy ran a deep-pull through public records. The Fulton County Clerk of Superior Court recorded lease memorandum surfaced a tenant termination right at year 7 — 18 months notice, six months penalty rent — not disclosed in the sponsor’s credit package. The EPA Envirofacts + Brownfields ACRES database flagged an active consent decree for chlorinated-solvent remediation on an adjacent 14-acre parcel held by the same SPV — a direct lien-exposure path. And 18 months of institutional-grade industrial comps showed South Fulton cap rates compressing back from 5.4% to 6.6%.
Recomputed value: $108M, not $124M — a 13% overstatement. True LTV at the requested loan: 74.6%, not 65%. The bank declined the structure as proposed, counter-offered $58.6M against the corrected basis, and required a tenant non-disturbance covenant at close. Credit committee confidence moved from yellow to dark green — because Heavy showed them where the risks actually lived.
Every number, click-to-source. The whole model is glass.
Sources: 11 verified · 0 below confidence threshold · last refresh 4h ago
Click any number. See the formula. See the inputs. See the source for each input. See the confidence score. See when it was last refreshed.
When a number lands on your screen, you don’t have to trust Heavy on faith. You can audit it down to the row of data it came from. Take it into IC. Hand it to LPs. Defend it under scrutiny. Every line carries its receipts.
A model that can’t explain itself can’t be trusted at scale. Heavy explains itself by default — the explanation isn’t a feature, it’s the architecture.
Type one line. Heavy returns the NOI, current value, and value-add potential — without a rent roll, without a T-12, without an OM.
No user-supplied data required. Heavy retrieves what it needs from public records, municipal portals, paid feeds, and primary sources — and returns a defensible valuation in seconds. Built for instant deal triage, opportunistic sourcing, and screening hundreds of targets a day before any analyst touches a model.
* Accuracy benchmarked against 135 properties in the United States, measured against subsequent verified appraisals and transaction outcomes.
Other AI is a tool you operate. Heavy is a team that operates for you.
A tool gives you intermediate output. You still have to think, verify, reconcile, decide. A team gives you finished work — modeled, sourced, signed, decision-ready. With ChatGPT or Claude, you operate the assistant. With Heavy, the system operates for you. Hand it the task. The task is done.
Titleman Light
The full institutional stack. Not a feature list — a replacement for the analyst floor.
Every asset class. Including the ones that haven’t been named yet.
If it produces cashflow or appreciates land value, Heavy can underwrite it. We’ve modeled assets that didn’t have a category three years ago and will have a sub-category in three years.
Meet Compass. The market discovery layer inside Heavy.
Real estate alpha does not live in the markets every fund already underwrites. It lives in the ones nobody bothered to model — the county where the supply pipeline collapsed, the MSA where institutional capital just fled, the submarket where infrastructure flipped on last quarter. Compass scans 3,143 US counties and 600+ international markets, by every asset class, against 5,000+ data sources — and surfaces the cells where institutional IRR is actually living. Not the consensus top twenty. The right one — for your asset class, your strategy, your mandate.
- 3,143 US counties + 38,000+ Census tracts + 600+ international markets ingested at primary-feed granularity.
- Every asset class fitted to every county — multifamily, industrial, retail, office, hospitality, healthcare, life-sciences, data-center, self-storage, land, niche.
- 62,860+ cells scored per US cycle on eight signal families, weighted to your mandate.
- Submarket drilldown: county is the entry point, corridor / tract / catchment is the trade.
- Weekly delta refresh, monthly deep re-rank. Every output signed, sourced, reproducible.
- SupplyPipeline drought, pre-leasing velocity, replacement-cost basis vs trading basis.
- DemographicNet IRS / USPS migration, QCEW wage growth, household formation, key-cohort momentum.
- Capital flowsRCA bid & ask volume, capital-flight signals, foreign vs domestic flow, sponsor concentration.
- InfrastructureNew port, rail, fiber, transmission capacity. Federal IRA / CHIPS allocations, county-level.
- Regulatory & basisRecent zoning amendments, OZ catalysts, abatements, cap-rate spread vs comparable-risk benchmark.
“Northern Virginia is the data-center market.” Loudoun County is the global default. Vacancy 1.2%, $200/kW, $1,100/sqft basis. Every hyperscaler is there; every fund is bidding into it. The trade is to get in, not to find. Headline projected IRR: 12.0%.
Cheyenne, Laramie County. Four reinforcing signals: 230kV transmission surplus — 1.4 GW available capacity vs NoVa negative; 92% lower power costs ($0.038/kWh vs $0.41/kWh blended); federal-land tax abatement via Wyoming Business Council; sub-50ms NYC latency via existing dark-fiber routes. Land basis $0.84/sqft vs Loudoun $185/sqft. Projected IRR 24.1% · spread +1,210bps · confidence 91%.
Meet Plays. The strategy & risk discovery layer inside Heavy.
Real estate is not one asset class. It is more than a hundred — multifamily, office, warehouse, land, hospitality, self-storage, data centers, life sciences, IOS, marinas, cell towers — and each one has its own deep playbook for unlocking value. Standard LLMs check the same three plays on every property they ever see: Hold, Flip, BRRRR. Plays runs the full applicable library against the asset in front of it — every eligible value-add strategy on the upside, every red flag the OM was hoping you wouldn’t check on the downside. In parallel. On every deal Heavy touches.
- Property fingerprinted across 42 dimensions — 3,200+ data points per address.
- One playbook loaded for the asset class. No cross-contamination.
- Every eligible play survives a hard-constraint gate — base zoning, lender covenant, capital horizon.
- All survivors underwritten in parallel. No top-N truncation. Completeness is the product.
- Compatible plays stacked into combined-strategy scenarios. Every output signed and reproducible.
- TitleLiens, easements, ROFRs, ground-lease resets — checked against the recorded chain.
- EnvironmentalPhase II, vapor intrusion, UST, floodplain, WUI — primary registries, not summaries.
- RegulatoryNon-conformance, illegal units, pending downzonings, STR ordinances, rent overlays.
- Lease & tenantTermination rights, co-tenancy, unfunded TI, below-market clauses, anchor credit drift.
- Capital & marketRefi risk, covenant trips, cap decompression, supply pipeline, insurance withdrawal.
Three plays: Hold for cash flow at 8.4% projected IRR. BRRRR via interior renovation. Flip at year three. Recommended: Hold-and-stabilise. Headline IRR target 11–12%, “consistent with Phoenix MSA garden-style benchmarks.”
23 plays scanned. The parcel — flagged by the fingerprint as R-3 zoning with a 2023 Maricopa County ADU code amendment — is eligible for ADU addition on 22 ground-floor units, stacked with sub-metering and Section 8 conversion on 12 units. The top three Plays-ranked strategies do not appear anywhere in the standard LLM response. Combined IRR: 19.4%. Equity uplift: $8.6M.
Meet Scout. The off-market machine inside Heavy.
Heavy doesn’t wait for deals to land in your inbox. Drop a screenshot of any map — one block, one corridor, one entire metro — and Scout reads every visible address, runs a full underwrite on each, and surfaces the assets where 19–24% IRR is hiding in plain sight. Off-market. Mispriced. Mismanaged. Distressed. Before they ever reach a broker channel.
- You drop a screenshot of any map — satellite, street grid, paid platform, hand-drawn polygon.
- Scout reads every visible address and pulls each into the verification stack.
- Twelve signal families run in parallel: pricing, rents, vacancy, debt, distress, owner, capex, regulatory.
- Each candidate gets a full underwrite, a recommended bid, and a defensibility-graded IRR thesis.
- You receive a ranked shortlist of off-market deals — with bid prices — ready for IC.
- InvestorsStabilised buys, value-add, distressed, secondary LP positions.
- DevelopersLand at prior use, recently upzoned parcels, stalled projects, OZ-clock plays.
- LendersRefi-pipeline borrowers, NPL pools, CMBS-risk assets, distressed-debt origination.
- Owners & landlordsAdjacent assemblages, sub-market expansion, portfolio adds.
- GeographyUS, Canada, Europe, GCC, LATAM, Türkiye, Japan, Australia — anywhere a parcel exists.
We’re giving Heavy arms. And then feet.
Until now, every AI investment tool has been disembodied — confined to whatever data the internet happened to expose. Heavy is leaving the screen. It is reaching into the physical world to verify what cannot be verified online.
- You select the comp set or competitor list inside Heavy.
- Arms generates a calling & email plan and runs it autonomously.
- Brokers, agents, managers respond. Conversations transcribed live.
- Heavy parses transcripts into clean fields — asking rent, concessions, lease terms, vacancy, rumored exits.
- Result lands inside your model in hours. You pay tokens for what was actually fetched.
- Heavy selects a profile that fits the asset (income, family stage, location, language).
- The profile is dispatched to view the property as a buyer or tenant — anonymous to the seller.
- On-site, they collect real rent, real concessions, real condition, real contract terms.
- Photos, voice notes, signed forms, leasing materials — all uploaded into your live deal file within 48 hours.
- Shopper is paid. Heavy doesn’t pretend it knows. It actually went and looked.
A two-sided economy built into your underwriting workflow.
Nonein practice.
Heavy replaces the full workflow. There is no part of pre-acquisition underwriting, asset management, or portfolio surveillance the system does not handle.
Your investment engine.
The core system that runs your business.
Heavy is sold by mandate, not by license. We deploy alongside one investment platform at a time, configured to your strategy.
Typical deployment, four to six weeks. Onboarding includes ingestion of your historical deals, templates, and underwriting standards.