Previous Projects

MICHAEL WIBRACHT

Founder & Principal, AllDark.Horse

San Antonio, Texas

Twenty-five years underwriting, capitalizing, and building real estate. Now building the software that work should have had.

OVERVIEW

Michael Wibracht is a real estate developer turned AI solutions architect. Over 25 years he co-founded and led 210 Development Group and Mission DG, where he served as Co-Founder, Managing Partner, and President and drove the strategy, capitalization, and execution of market-rate and affordable multifamily, mixed-income and mixed-use development, for-sale residential, adaptive reuse, urban infill, and complex public-private partnerships across Texas — more than $1 billion in construction and development, and over 6,000 residential units delivered or in the pipeline.

He holds a BBA in Management Information Systems from Texas A&M University, College Station. That combination — a systems education and a quarter century of closing capital-intensive, multi-source transactions — is the whole premise of what he does now.

Through AllDark.Horse, his AI solutions studio, Michael designs and operates purpose-built AI financial and design tools for real estate development finance across the full product spectrum: market-rate rental, affordable and mixed-income, for-sale residential, adaptive reuse, and commercial and mixed-use. He is sole inventor on ten United States provisional patent applications filed in 2026, comprising 308 claims, held by AllDark Horse LLC. The tools are not adapted from generic software. They were specified by someone who has personally sized the loan, defended the basis, and signed the guaranty.

WHY ALLDARK.HORSE EXISTS

Development finance is not hard because the math is hard. It is hard because the math lives in forty places at once — a lender's term sheet, an equity pricing letter, a construction budget, a rate that moved this morning, an architect's drawing that has not been reconciled to the rent roll or the price list — and because a single assumption drifting out of date can quietly invalidate an entire capital stack weeks before anyone notices.

That is true of a 300-unit market-rate podium deal, a for-sale townhome community with a unit-release schedule, and a four-layer LIHTC historic adaptive reuse alike. The sources differ; the failure mode does not.

Adaptive reuse adds a second problem on top of the first. The conversion of obsolete office buildings into housing is limited far less by architectural imagination than by a small number of physical facts about the existing building that are, today, discovered late and by hand: where a new plumbing riser can actually be cored through the slab, how deep a habitable room can run from the glass before it stops being lightable, whether the as-built corridor still meets egress distance, and what the building actually measures today versus what the record drawing says it measured in 1974.

Generalist AI platforms solve neither problem. They do not know that 221(d)(4) amortizes over 40 years and prices differently for market-rate than for a high-rental-assistance property, that the applicable fraction is not the same as the qualified basis, or that the statutory instrument a project requires changes the moment tenure changes from rental to for-sale. And they have no way to measure a building. Specialists in this industry, meanwhile, have historically been armed with spreadsheets and institutional memory.

AllDark.Horse is built at that intersection: software with the domain judgment of an operator encoded into it, and the geometry of the actual building underneath it.

DEVELOPMENT TRACK RECORD

- $1B+ in construction and development overseen

- 6,000+ residential units delivered or in development

- $405M total development investment

- $140M in historic adaptive redevelopment

- $114M LIHTC eligible basis

- $31M in Federal Historic Tax Credit equity

Product experience spans market-rate and workforce multifamily, affordable and mixed-income rental, senior housing, for-sale residential, mixed-use, and historic adaptive reuse — garden, wrap, podium, mid-rise, and conversion product.

Transactional experience spans HUD 221(d)(4) and 223(f) across market-rate, middle-income, affordable, and rental-assisted tiers; Fannie Mae DUS and Freddie Mac; bank and bridge construction debt; LIHTC (4% and 9%); Federal and State Historic Tax Credits; New Markets Tax Credits; and joint-venture and preferred equity structures.

INTELLECTUAL PROPERTY

Ten U.S. provisional patent applications, filed March through September 2026, comprising 308 claims. Michael Wibracht is sole inventor on each. The portfolio is held by AllDark Horse LLC, with assignment of the earlier applications recorded with the USPTO in August 2026 and the balance in process. The ALLDARK.HORSE and DIGITAL TWIN ENGINE trademarks are pending in Class 042.

Digital Twin Engine — filed March 31, 2026 (17 claims)

Automated multi-layered capital-stack underwriting, rate-driven re-underwriting of the full pipeline on every rate refresh, 0-100 viability scoring with tiered alerting, bond allocation tracking, and predevelopment schedule generation.

Nirmata Core — filed April 8, 2026 (34 claims)

Automated building footprint extraction, residential unit layout generation, code compliance enforcement, and dual-format cost estimation across four independent claim groups.

The Kasati — filed April 23, 2026 (29 claims)

Authenticated multi-source real estate opportunity discovery with cross-source entity fusion and layered deduplication.

Digital Twin Engine II — filed April 25, 2026 (22 claims)

Automated geographic eligibility determination for tax credit boost and Opportunity Zone status, with real-time capital stack re-underwriting on determination.

OM Pipeline — filed July 1, 2026 (22 claims)

Automated pipeline converting unstructured real property offering memoranda into code-constrained development programs, architectural geometry, financial models, and investor artifacts.

As-Built Registration — filed July 31, 2026 (22 claims)

Registration of a generated residential unit layout to an existing building's as-built architectural drawing by computer vision, with model-to-as-built reconciliation and Historic Tax Credit qualified-rehabilitation-expenditure segregation.

Constraint-Scoring & Scheme Selection — filed August 7, 2026 (38 claims)

Geometric classification of rehabilitation expenditures against a registered as-built building envelope, with constraint-based verification and selection among generated layouts.

Tenure — filed August 13, 2026 (49 claims)

Disposition-path-dependent generation, classification and ranking of residential layouts for adaptive reuse of existing buildings of any age — across rental, subsidized rental, and for-sale tenures — with jurisdiction-parameterized code compliance, a federally insured financing overlay, and generation of the statutory instrument required by the elected tenure.

Nirmata Core II — filed August 21, 2026 (22 claims)

Residual-gated registration of a generated layout to an as-built structural grid, riser-first unit subdivision, and geometry-driven dual-format cost propagation.

Nirmata Core III — filed September 4, 2026 (53 claims)

Structural-coring feasibility gating of new plumbing risers, a facade-derived lightable-depth envelope, egress-distance gating, and pro-forma-inverted layout generation, with construction of a measurement-derived reconciled as-built record carrying per-dimension provenance, source adjudication, volumetric envelope classification, and uncertainty-gated propagation.

The earliest application's non-provisional conversion window runs to March 31, 2027.

THE PLATFORMS

The platforms are asset-class agnostic by design. Affordable and adaptive reuse deals are the hardest case — the most capital sources, the most compliance constraints, the most ways to be wrong — so the systems were built to that standard and run market-rate and for-sale product on the same engine.

Digital Twin Engine — the pipeline that re-underwrites itself

The problem: a deal underwritten in March is a work of fiction by June. Rates move, spreads move, sources open and close, and nobody re-runs the model until something has already gone wrong.

What it does: Digital Twin Engine is the unifying platform layer — live pipeline intelligence with FRED API rate integration, automated daily re-underwriting, viability scoring on a 0-100 scale, and tiered email alerts when a deal's position materially changes. HUD spreads are calibrated against the 10-year Treasury and MIP is carried flat at 25 basis points. A single address entry triggers full site classification — Qualified Census Tract, Difficult Development Area, and Opportunity Zone status determined automatically against Census and HUD endpoints — and cascades the result through every downstream calculation, applying the 130 percent basis boost where it is triggered. A FastAPI REST backend and React dashboard expose sensitivity analysis and full capital-stack modeling: conventional and HUD-insured debt, construction and bridge facilities, joint-venture and preferred equity, and a layered soft-source gap filler covering deferred developer fee, PBV, HOME, CDBG, TDHCA, local gap/TIF, FHLB AHP, and NMTC. Bond allocation and state credit pool availability are tracked alongside. Containerized, with automated rate cycles, weekly Excel exports, and SSL-secured multi-tenant deployment.

Why it matters: the portfolio tells you when it has changed, instead of waiting to be asked. That applies to a market-rate refinance watching its rate lock window as much as it does to a nine-source affordable deal.

Nirmata Core — the building, measured

The problem: the gap between a set of construction documents and a defensible development budget is measured in weeks of manual takeoff. Worse, on an existing building, the drawings are frequently wrong — and the facts that kill a conversion are physical ones nobody checks until construction documents.

What it does: Nirmata Core ingests DWG/DXF natively, converting construction documents into structured building geometry with no manual takeoff, and registers generated layouts back onto the building's actual as-built structural grid by computer vision — gated on registration residual, so a layout that does not truly fit is rejected rather than reported as fitting. Where record drawings are missing or unreliable, a three-dimensional measurement dataset of the building as it stands today is reconciled against the record drawing rather than substituted for it, producing an as-built record in which every dimension carries the source, method, and tolerance by which it was obtained, and in which divergence between sources is adjudicated on the record.

On that geometry the platform runs the gates that decide whether a conversion is real: whether a proposed new plumbing riser can physically be cored through the existing slab, and with what clearance; how deep a habitable room can run from the facade before it stops being lightable; whether as-built corridor travel distance still satisfies egress; and what the rehabilitation actually costs, classified geometrically against the registered envelope so that Historic Tax Credit qualified rehabilitation expenditures are segregated by where the work physically occurs rather than by invoice caption.

Layouts are then generated riser-first and pro-forma-inverted — worked backward from the financial outcome the deal has to produce — across rental, subsidized rental, and for-sale tenures, with jurisdiction-parameterized code compliance and automatic generation of the statutory instrument the elected tenure requires. Cost estimates output a development cost schedule and a CSI MasterFormat division breakdown from a single computation.

Why it matters: the architecture, the building's real geometry, and the capital stack stop being three separate exercises reconciled by hand — whether the output is a rent roll or a price list.

The Kasati — deal sourcing with provenance

The problem: the opportunities worth finding are buried in government parcel records, federal overlays, and foreclosure notices that do not speak to each other, and most tools that claim to aggregate them are scraping their way toward a cease-and-desist.

What it does: The Kasati is an authenticated multi-source opportunity pipeline surfacing development and redevelopment targets — infill sites, underused and vacant buildings, conversion candidates, and portfolio acquisitions — from government parcel records, federal overlays, tax foreclosure notices, and market listings. Records are fused across sources via normalized-address fingerprinting with per-attribute source provenance preserved, layered deduplication applied, and results scored against a multi-dimensional viability rubric tunable to market-rate, for-sale, or affordable criteria.

Why it matters: every field in a Kasati record can be traced to the source that produced it — and the platform operates exclusively on authenticated APIs and user-initiated data transfers. No scraping. No terms-of-service violations.

The OM Pipeline — from offering memorandum to investment committee

The problem: a broker's offering memorandum is a PDF written to sell, not to underwrite. Turning one into a development program, a massing, a model, and an IC package is a week of work — repeated on every deal that turns out not to pencil.

What it does: the OM Pipeline converts unstructured offering memoranda into code-constrained development programs, architectural geometry, financial models, and investor artifacts, automatically.

Why it matters: the deals that do not work fail in an afternoon instead of a week, which is what makes it possible to look at all of them.

FOUNDATIONAL FINANCIAL MODELS

The platform layer sits on a base of individually validated models:

- 10-Year Treasury Predictor with HUD Pricing — automated daily FRED pull, three-model forecast engine (momentum, mean reversion, volatility-adjusted), calibrated HUD 223(f) and 221(d)(4) spreads, weekly email delivery

- HUD 223(f) Loan Sizer — DSCR-constrained mortgage sizing for acquisition and refinance across market-rate (87% LTV / 1.15x), affordable and middle-income (90% / 1.11x), and rental-assisted tiers, with current MIP at a uniform 25 bps per FR-6522-N-02, plus LTV caps and debt yield analysis

- HUD 221(d)(4) Loan Sizer — new construction and substantial rehabilitation: 40-year amortization, tier-specific DSCR and loan-to-cost constraints, and construction-period interest modeling

- LIHTC Financial Model — full 4% and 9% underwriting with eligible basis calculation, applicable fraction, 10-year credit stream, investor pricing, Texas State LIHTC (HB 1058) layer, and TDHCA-format Development Cost Schedule

- Fannie Mae Small Loan Sizer — side-by-side Freddie Mac SBL and Fannie Mae comparison with LTV/DSCR constraints and rate lock analysis

HOW THE WORK IS BUILT

Five commitments govern every platform AllDark.Horse ships:

Provenance over convenience. Every data point carries its source. Fused records preserve per-attribute origin, and every reconciled dimension carries the method and tolerance by which it was obtained, so a number can always be defended, not just displayed.

Authenticated sources only. Government APIs and user-initiated transfers. No scraping, no gray-area acquisition, nothing that creates exposure for the client downstream.

Measured, not assumed. On an existing building, the drawing is a claim and the measurement is the evidence. Where the two diverge, the divergence is adjudicated and recorded rather than quietly resolved.

Say when you cannot tell. A corridor measured at forty-four inches against a forty-four inch limit passes a binary test and fails an honest one. Where measurement uncertainty makes a determination indeterminate, the platform reports it as indeterminate and names the field measurement that would resolve it — rather than returning a confident answer the evidence does not support.

Domain fidelity first. A model that is elegant and wrong about a DSCR tier or an applicable fraction is worthless. Correctness against the actual regulatory and credit framework precedes everything else.

WORKING WITH ALLDARK.HORSE

For developers and builders — platform access, custom model development, and pipeline intelligence built around an existing portfolio and deal criteria, across rental, for-sale, and mixed-use product.

For lenders, syndicators, and capital partners — underwriting and screening tools calibrated to specific credit boxes, program requirements, and asset classes.

For principals and joint-venture partners — direct development expertise on ground-up multifamily, adaptive reuse, historic credit, and layered transactions.

TECHNICAL DETAIL

Stack: Python (FastAPI, pandas, NumPy), React/JavaScript, REST API design, Docker containerization, OAuth 2.0 authentication, FRED API integration, automated email delivery, nginx reverse proxy with SSL, cron-based scheduling, PostgreSQL, libredwg and ezdxf for CAD pipeline ingestion, computer-vision registration of generated geometry to as-built drawings, three-dimensional measurement ingestion, and full-stack cloud deployment on DigitalOcean/Ubuntu infrastructure.

Domain specialization: AI-powered underwriting and design automation across market-rate multifamily, workforce and middle-income housing, affordable and mixed-income rental, for-sale residential, senior housing, mixed-use, and adaptive reuse. Financing structures include HUD 221(d)(4) and 223(f) across all property tiers, Fannie Mae DUS and Freddie Mac SBL, bank and bridge construction debt, LIHTC (4% and 9%), Federal and State Historic Tax Credits, New Markets Tax Credits, Project-Based Vouchers, and layered public capital including HOME, CDBG, TDHCA soft debt, FHLB AHP, TIF/TIRZ, and seller financing.

PROFESSIONAL STANDING

Mr. Wibracht maintains full eligibility to participate in federally related and regulated transactions and is not suspended, debarred, or restricted from engaging in projects involving institutional or government-related capital.

CONTACT

AllDark.Horse

San Antonio, Texas