An AI-powered SMB funding engine: faster, fairer, and more transparent than the incumbent it is benchmarked against, with modular state-law overlays. Objective, as written: build a 100× faster, fairer, and more transparent SMB-funding engine than Closer Capital, on the proprietary Neuruh Execution Stack.
| Dimension | Closer Capital (baseline) | LENDURUH Advantage |
|---|---|---|
| Application | Web form; human review; 24-h follow-up. Qualify needs: 3+ mo in business, 600+ FICO, $10K+/mo revenue. | Plaid-powered bank link + instant KYC ⇒ soft pull → < 120 s conditional approval. OAuth/Plaid/Stripe with OFAC + device-risk fingerprint. |
| Underwriting | Hard rules: FICO 600+, > $10K/mo, > 3 mo bank history. | Multivariate AI score (cash-flow velocity, seasonality, real-time sales, POS feeds, payroll, social reputational graph) → qualifies thin-file & gig-economy firms. |
| Product Set | MCA, term loans. | Revenue-based financing, dynamic credit lines, equipment leases, on-chain asset-backed notes. |
| Turnaround | “Lightning fast” (claims). | < 2 h funding via API wire trigger; < 2 h disbursement (instant-ACH / RTP*) via rules-driven wire API. |
| Transparency | Marketing-heavy, generic FAQs, opaque rates. | Live risk dashboard, APR disclosed to the penny, payoff simulation, in-product SHAP-based explainer. |
| Moat | Sales brand around Ryan Stewman. | Proprietary Neuruh Execution Stack: Underwrite-Agent, Fraud-Agent, Compliance-Agent, Customer-Success-Agent, all self-optimizing, continually RL-tuned on portfolio outcomes. |
*RTP = real-time payments via FedNow / The Clearing House. Baseline facts sourced in the document from closercap.com.
The document includes full build instructions for a minimum viable slice: a FastAPI micro-service that exchanges a Plaid public token, pulls a year of transactions, extracts cash-flow features, scores them with a trained model, and answers on POST /apply. Its decision rule is the one the simulator in section 03 runs: score > 0.65 → APPROVED, otherwise REVIEW; risk bands A above 0.80, B above 0.65, C below. Expected test response in the document: {"score": 0.72, "decision": "APPROVED", ...}.
The document's underwriting thesis: no FICO gate. The score is multivariate, built from cash-flow behavior, so thin-file, gig-economy, and immigrant-founded firms can qualify on how the business actually moves money.
Demo formula (stand-in for the trained model): score = 0.45 × revenue factor + 0.25 × balance factor + 0.30 × stability factor − negative-day penalty. Decision and band thresholds are the document's own: APPROVED above 0.65, band A above 0.80, B above 0.65, C below. The contribution bars below play the SHAP explainer's role: every decision shows its reasons.
All decisions are Reg B / ECOA + state fair-lending logged for downstream examiners. The ROE's five components:
| Component | Functionality | Tech notes |
|---|---|---|
| State Rule Loader | JSON schema per state: licensing class, APR cap, fee cap, prohibited clauses (COJ), disclosure timelines. | Postgres → cached in Redis |
| Field-Gatekeeper | Masks or exposes application fields as required by state (e.g., NY §23-2.4 requires prior MCA disclosures). | Next.js renders conditionally via feature flags |
| Usury Guard | Blocks pricing above statutory cap and routes to a compliant product (e.g., revenue share instead of LOC). | LangGraph node before Term-Sheet generation |
| Broker-of-Record Resolver | Confirms LENDURUH holds the proper license or auto-assigns a licensed partner in-state. | Middesk Licenses API + partner registry |
| Audit Trail Generator | Packages borrower-state snapshot, model explanation, and policy references into an immutable PDF. | PDFKit → S3 → Glacier; hash anchored to public chain nightly |
The state-law overlay module (v0.9) plugs in per state:
| Overlay | Key inputs | Runtime action |
|---|---|---|
| Usury & Fee Caps | APR cap %, late-fee max | Block pricing or auto-route to revenue-share product |
| Licensing | Lender, broker, sales-finance | Ensure agent-of-record + bond filing present before offer |
| Confession-of-Judgment Ban | COJ allowed? Y/N | Strip COJ clause from docs in states that disallow |
| Privacy & Data | CCPA / CPA / SHIELD | Trigger additional disclosure banners & opt-out options |
| Lien & UCC Rules | Filing deadlines | Auto-file or auto-escalate to counsel based on state |
Set a state's APR cap and a proposed APR. The guard is a hard gate, not a warning: above the cap, pricing is blocked and the file auto-routes to a revenue-share product, which carries no APR. Cap values here are yours to set; the document stores the real tables per state in Postgres.
The Lender Intelligence Layer crawls live lender data, normalizes it into the LenderGraph, and keeps it fresh:
| Sub-agent | Data source | Refresh | Output |
|---|---|---|---|
| LenderCrawler-Agent | Public rate pages, SEC filings, CFPB 1071 dataset, state NMLS feeds, FDIC call reports | 2× daily | Raw term-sheet HTML/JSON (diff-store) |
| Parse-Clean-Enricher | Regex & ML entity extract; normalizes APR vs factor; joins licensure & coverage states | Real-time via Kafka | Canonical lender record |
| LenderGraph DB | Neo4j graph: Lender, Product, State-License, Risk-Appetite nodes | On ingest | Queryable rankings |
| RatePredict-Agent | XGBoost → predicts next-week rate drift per product to advise timing | Nightly retrain | Rate-movement score |
The document's scoring factors and weights: Product Fit 35%, Net Cost to Borrower 25%, Speed SLA 15%, Flexibility 15%, Reputation 10%. Three sample lender records (the document's tuning seed calls for 50+ lender/product combos) are ranked live. The borrower can re-run with different priorities, as documented: switch the preset and watch the Top-3 reorder. The winner's strongest factor is its “Why This Is Best” explainer.
| # | Lender (sample record) | Fit | Cost | Speed | Flex | Rep | Score |
|---|
% of daily card/POS sales until a payback multiple is reached; resets dynamically with seasonality.
LOC that auto-ratchets up/down on weekly data pulls; pricing floats on risk-score delta.
12–48 mo lease with buy-out slider; assets tokenized for secondary sale to yield desks.
Reg D SPV; loan fractionalized as fixed-yield token (whitelisted wallets).
Real-time underwriting of eligible invoices via Stripe AR or QuickBooks.
If the Neuruh risk score is below threshold, auto-route to an SBA micro-loan or CDFI partner.
The grid promises a borrower-facing payoff simulation with APR disclosed to the penny. This sandbox does the arithmetic for a revenue-advance style deal: total payback, weekly and daily remittance, cost of capital, and the simple annualized equivalent of the factor over the term. Illustrative arithmetic only, not an offer.
Asks: EIN, state of org, UBO percentages, foreign owners?
Decision purpose: KYC/AML, state-law routing.
| Block | Purpose | Core data / logic |
|---|---|---|
| Identity & Fraud Core | KYC/AML, OFAC, synthetic-ID check | Middesk + Socure + device fingerprint (fingerprintjs) |
| Cash-Flow Graph | Underwrite non-FICO health | Plaid Auth/Transactions, Stripe Balance, QuickBooks P&L, Shopify orders, UPS/FedEx scans |
| Reputational Graph | Reward honest operators | LinkedIn org graph, Trustpilot, public suits/UCC liens, GitHub/social for tech firms |
| Policy Engine | Constraints that must fire before approval | CFPB MLA/SCRA flags, Reg Z disclosures, FDIC AML rules |
| Explainability Layer | Borrower-visible & regulator-auditable | SHAP visual + narrative; stored with loan file (S3 → Glacier) |
| Self-Optimizing Loop | Edge-case capture & RL tuning | Portfolio KPIs → feedback to XGBoost/RL module nightly |
| Security area | Control |
|---|---|
| Data-at-Rest | AES-256; envelope keys (KMS) per borrower |
| Data-in-Transit | mTLS between micro-services |
| ML Governance | Model cards, bias drift monitor, SOC 2 change-control log |
| Reg-Tech | Automated CTR/SAR draft via Middesk; OFAC delta-scan every 24 h |
| Audit Trail | Immutable LedgerDB; SHA-256 hash anchor to public chain every midnight UTC |
| Window | v0.9 Build-Out (federal layer first) |
|---|---|
| Week 0–2 | Spin up dev cluster, integrate Plaid sandbox, baseline XGBoost model on synthetic cash-flow. |
| Week 3–4 | Deploy Decision Orchestrator; stub Compliance-/Fraud-Agents; ship Q&A wizard v0.1. |
| Week 5–6 | Launch Funding Rails (Dwolla sandbox + RTP); live pilot in Delaware (usury-friendly). |
| Week 7–10 | Add first five overlay states (TX, CA, FL, NY, IL) + SOC 2 Type I audit kickoff. |
| Week 11–12 | Tokenized Note MVP (Polygon PoS, KYC-gated); open partner marketplace for off-grid approvals. |
| Sprint | v1.1 Delta Deliverable | Success metric |
|---|---|---|
| S-1 | LenderCrawler-Agent MVP (10 lenders, 2 states) | > 90% term-sheet field accuracy |
| S-2 | LenderGraph DB deployed; basic rank algorithm | Top-3 list renders < 400 ms |
| S-3 | ROE integrated → field-gate + usury guard | Zero compliance warnings in internal QA |
| S-4 | Closed beta in DE + TX; 20 live deals | ≤ 2 h median funding time |
| S-5 | Crawler to 100 lenders, all 50 states; fairness & drift monitors live | Disparity index < 5% across protected classes |
Part 1 of the source is a five-minute response-video playbook: neutralize the “AI-is-snake-oil” narrative without naming its source, demonstrate execution, convert viewers into qualified leads or investors. Delivery: corporate-confident, slightly irreverent, data-heavy, ~160 wpm, under 5:30.
| Segment | Time | Core message | CTA |
|---|---|---|---|
| 1 · Hook | 0:00–0:15 | “Everyone's suddenly an ‘AI expert.’ Turning AI into cash-flow is a different universe.” | None · hold suspense |
| 2 · Reality Check | 0:15–1:00 | 99% of the industry sells prompt templates. Neuruh ships working agents that replace payroll and print profit. | Lower-third: Neuruh AI Execution Engine |
| 3 · Proof-Stack | 1:00–2:15 | Live prototype demo plus the script's proof points: beta users, a signed pilot, early pipeline, pending LOIs. | “Type DEMO below” |
| 4 · Myth-Busting | 2:15–3:00 | Three myths refuted with metrics and deployments: GPTs are toys; AI is copy-paste code; small teams can't win. | Link card: Read Case Study |
| 5 · Transformation Demo | 3:00–4:30 | A manual funding workflow, then one click: the Funding Agent assembles docs, underwrites, issues a term sheet in 2 minutes. | “Message AUTOMATE to pilot” |
| 6 · Call-Out Without Name-Drop | 4:30–5:00 | “If your business model depends on calling AI ‘snake oil,’ maybe you're still in the horse-and-buggy business. We're building the highway.” | None |
| 7 · CTA | 5:00–5:20 | “Want operating leverage instead of buzzwords? Book the seven-minute strategy call.” | Seven-Minute Strategy Call |
Segment 3's specific figures (user counts, pilot name, pipeline and LOI amounts) appear in the source as script copy. They are the document's claims, not verified results, and are not restated as numbers here.