# Enginuity: Foundation & Architecture (S4 – S9) **Student:** Max Bellomy **Course:** AI-Accelerated Entrepreneurship Practicum (Spring 2026) **Venture:** Enginuity — AI decision intelligence for custom manufacturers under $150M. --- ## S4: Knowledge Architecture Doc & "What We Don't Know" **1. The Domain Reality** Enginuity operates in the complex data environment of custom manufacturing. The core problem is that the revenue vs. margin tradeoff is invisible in real time. To solve this, our Knowledge Graph must map the exact lifecycle of a custom part from estimation to factory floor execution. **2. "What We Don't Know" (Risk Analysis)** * **The "Napkin Drawing" Gap:** Customers often submit crude sketches. The AI cannot (and should not) translate a napkin directly into an engineering BOM. That requires human spatial reasoning. * **Phantom Floor Fixes:** If a welder fixes an engineering flaw on the floor but doesn't log the extra 2 hours in the ERP, the AI’s historical data is blind to that specific variance. * **Material Cost Volatility:** If the ERP's raw material cost table isn't updated weekly, the estimated margin on the SQ will be fundamentally flawed before the AI even evaluates it. --- ## S5 & S6: AI Reliability & Domain Risk Audit **1. Where Baseline AI Fails in Manufacturing** Standard AI models fail in custom manufacturing because they are inherently optimistic and lack grounding in physical constraints. If asked to quote a custom HVAC unit, a generic LLM will hallucinate a BOM and confidently predict a 45% margin based on industry averages, completely ignoring the specific factory's welding bottleneck or historical scrap rates. **2. Grounding the Enginuity AI** To ensure absolute reliability and prevent hallucination, Enginuity's AI is explicitly restricted from *creating* the initial Quote (SQ) or BOM. * **Human Domain:** The human technical estimator translates the napkin/engineering drawing into the initial SQ and labor routing. * **AI Domain:** The AI strictly *evaluates* the human's SQ against the Knowledge Graph of historical Work Orders (WO). **3. Kill Criteria (When the AI must escalate)** The AI is programmed to instantly flag and escalate any SQ where the true projected margin falls into the low single digits or goes negative. It also flags quotes that deviate by double digits from the historical 20-60% margin norms for that specific distributor vertical. --- ## S7: Persona Architecture Doc Enginuity utilizes a multi-agent "Pod" architecture. Instead of one AI trying to understand the whole business, specialized agents represent the distinct silos of a manufacturing plant, culminating in a master dashboard for the CFO. **1. The Technical Estimator Agent (The Auditor)** Intercepts newly drafted Sales Quotes (SQ). Compares the human estimator's BOM and labor routing against historical Work Order (WO) actuals to calculate the "True Projected Margin" before the quote is approved. **2. The Operations Agent (The Floor Manager)** Monitors live factory floor capacity and Work Order backlogs. Identifies shifting bottlenecks (e.g., Assembly vs. Custom Welding) and warns the Estimator Agent if lead times or labor costs will spike on incoming POs. **3. The Market Agent (The Strategist)** Analyzes the external value of the PO. Evaluates customer tiers, distributor verticals (e.g., Data Centers, Labs), and macro pricing trends. Determines if taking a low-margin job is worth the strategic volume. **4. The CIO Agent (The Synthesizer)** The ultimate interface for the CFO/Management. Aggregates conflicting data from Estimating, Ops, and Market into a single live Tradeoff Dashboard. Classifies active quotes into High, Medium, and Low Risk buckets, and provides direct, 3-part action recommendations (Summary, Status, Actions) to management. --- ## S8 & S9: Agent Logic Map & Context Architecture ### 1. The RAG Pipeline (S8) When a human saves a custom quote in the ERP, the Technical Estimator Agent triggers a 3-part retrieval mechanism: 1. **Historical Context:** Retrieves the last 50 Work Orders (WOs) that utilized the exact labor routings specified in the new quote. 2. **Strategic Context:** Retrieves the buyer's profile (industry vertical, lifetime value tier). 3. **Financial Context:** Retrieves the CFO's active margin floor target for that specific vertical. ### 2. The Custom Quote Approval Loop (S9 Logic Map) **Scenario:** A human estimator saves a custom HVAC quote in the ERP. * **Quote Size:** $120,000 * **Estimated Margin:** 24% * **Distributor:** Apex Climate (Tier-A, Data Center vertical). **Step 1: RETRIEVE** The agent intercepts the webhook and queries the Trinity Graph. *Result: Estimators consistently under-quote this specific welding routing by 15%. Target margin for Data Centers is 35%, absolute floor is 15%.* **Step 2: REASON (ReAct Prompt)** 1. **Analyze Baseline:** "The human estimated margin on this SQ is 24%." 2. **Apply Historical Correction:** "Graph data shows a 15% historical labor overrun on this specific WO routing. 24% estimated - 15% historical error = 9% True Projected Margin." 3. **Evaluate Against Target:** "The 9% True Margin is severely below the CFO's 15% absolute floor." 4. **Evaluate Strategic Context:** "However, the distributor is Tier-A, and the $120k volume serves the strategic Data Center vertical." 5. **Determine Action:** "Strategic volume overrides the hard stop rule, but the margin inversion presents severe execution risk. I will approve the quote for release to Sales, but flag it as 'High Risk/Review' on the CIO Tradeoff Dashboard so the CFO has visibility before the Purchase Order (PO) is received and converted to a Sales Order (SO)." **Step 3: ACT** 1. **ERP Update:** Changes SQ status to "Approved - Released." 2. **Dashboard Write:** Pushes the SQ to the CIO Tradeoff Dashboard under **High Risk - Labor Variance**, exposing the 9% true margin. 3. **Ops Alert:** Sends a silent pre-warning to the Ops Agent to anticipate a welding bottleneck if this PO is won. ### 3. S9 Memo: Why Context Changes Everything **To:** Professor Oliver Luckett **From:** Max Bellomy (Enginuity) **The Baseline AI Failure Mode** If you feed a standard LLM a $120,000 custom manufacturing quote and ask, “Should we approve this at a 24% estimated margin?” it will universally say yes. It operates in a vacuum. It doesn't know that the estimating team consistently under-quotes the welding labor on that specific custom part by 15%, or that the actual realized margin will drop to 9% once the job hits the floor. Relying on baseline AI would systematically destroy a company's profitability while aggressively growing its top line. **The Trinity Graph Advantage** Before our Technical Estimator Agent evaluates a Sales Quote (SQ), it executes a hard requirement to query the Knowledge Graph. It traces the full lifecycle from Sales Quote (SQ) to Sales Order (SO) to Work Order (WO). **Context Changes the Action** Because the agent possesses this structural context, instead of blindly approving a 24% estimated margin, the agent recalculates the *true projected margin* based on historical execution failure rates. It flags the impending labor variance on the CIO's Tradeoff Dashboard as "High Risk." Context is what makes the revenue vs. margin tradeoff visible before the quote ever leaves the building.