Launch of Industry-Facing PoC: “Estimating × Drawing Data Utilization” (Joint Project with a Manufacturing Client)

AIM Inc. has begun a proof of concept (PoC) with a manufacturing customer to cross-leverage existing forms, drawings, and evidentiary documents to improve both the accuracy and speed of estimating. This initiative respects the customer’s current workflows and formats, and is designed to lower operational load step by step.

Problems We Aim to Solve (Summary)

• Each site/department maintains forms with different column schemas and expressions, making cross-searching and reuse difficult.
• Information is scattered across handwritten notes, isometric drawings, and CAD.
• We need to ensure traceability of estimate rationales (timing, responsible person, source).

Project Overview

1. Natural-Language Cross Search Query in everyday terms like “item name,” “period,” or “owner,” and the system retrieves related information across sources.
2. “Variance Absorption” for Forms & Drawings Even without unified formats, the system normalizes semantically similar columns and terms into a common master.
3. Source Disclosure (Auditability) Every answer shows its references, avoiding back-and-forth confirmations on the floor.
4. Phased Rollout We first demonstrate value without changing existing formats, then expand cross-domain coverage and automation.

Technical Approach (Principles)

• Semantic Normalization: Combine LLMs with rule-based dictionaries to map differing column names/terms to common keys.
• Natural Language → Structured Query: Convert questions into an internal DSL/SQL-like form, achieving high recall via a hybrid of keyword matching and semantic search.
• Foundations of Drawing Understanding: Separate title block / tables / drawing regions and model text–geometry relationships as a graph.
• Agentic OCR: Handle diverse document layouts (POs, receipts, etc.) with a human-in-the-loop confirmation UI (double-check by design).

Intent of the principles: Fix the pipeline “field language → normalization → source-backed answer” to balance accuracy and availability.

Execution Plan (Phased)

• Phase 1: Secure data intake and create a minimal profile (define column semantics).
• Phase 2: Initial cross-search, source display UI, and lightweight OCR.
• Phase 3: Advanced handling of drawing symbols/annotations and connection to business outputs such as bills of materials.
• Phase 4: Holistic evaluation of accuracy, speed, and operations, and agreement on extension items.

Security & Operations

• Operate per customer policy, choosing among public cloud / private network / access controls.
• Enforce least privilege and minimal scope for data access; retain operation logs.
• Public disclosures remain within customer-approved bounds; no customer-identifying specifics are handled.

Anticipated Outcomes (Illustrative)

• Shorter time for search and reconciliation related to estimating.
• Lower communication overhead between field and back office by presenting supporting sources.
• Reduced data entry via semi-automatic extraction from drawings and evidentiary docs. Specific figures will be measured per the customer’s evaluation process and may be published after approval.

What’s Next

We will share PoC progress and results within the scope approved by the customer. For inquiries or demo requests, please contact us via the inquiry form.

This article contains no information that could identify any specific company (e.g., company names, dates, internal structures, personal names, or specific KPIs). The content is generalized, and actual implementation and operations follow the agreements with each customer.