Modernization and Intelligent Document Processing for land surveying firm

Business Context

A land surveying firm faced a regulatory requirement to digitize tens of thousands of historical survey records. The records varied in format based on region and era, requiring domain expertise to process correctly.

The firm was working with a limited budget due to timing constraints, and an approaching compliance deadline added pressure to find a workable solution quickly.

Problems

High processing time per document. Manual data entry required 2 to 4 minutes per survey record. Given the volume of documents and the regulatory deadline, this throughput was insufficient.
Non-standard document formats. Land survey records differ based on when and where they were created. Processing required understanding of regional conventions and historical naming patterns.
Partial budget. A previous software vendor had started but not finished developing an application, consuming budget that would otherwise be available for the new initiative.

Solution

Deciding to Pivot or Exit

The client was considering canceling the project due to budget constraints and timeline pressure. However, in early 2023, commercial AI capabilities had begun to demonstrate potential for document processing tasks. The client proposed a workflow that would leverage AI to extract data from survey images.

The technical feasibility of this approach was uncertain. I proposed building a proof of concept to validate whether AI could extract the required fields from land survey documents with acceptable accuracy.

Building the Proof of Concept

Working with the client's domain experts, I created a simple dataset that mapped what kind of data should come from each document, and how that data should be used.

I tested several extraction approaches against this dataset. The approach that performed best achieved approximately 85% accuracy in extracting key fields from previously unseen documents.

I recorded a video demonstration and presented it to the client's board. The board approved the project based on the demonstrated results.

Defining the Roadmap

The project scope expanded from simple form entry to AI-powered document processing. However, the original constraints remained: the regulatory deadline had not changed, and budget remained limited.

I defined a minimal roadmap to prioritize getting staff productive as quickly as possible:

User authentication and document upload
AI-assisted processing with immediate progression to the next document
Review workflow for staff to approve or correct extracted data
Export of approved records

The timeline from approved proof of concept to production application was approximately two months.

Building the MVP

I structured the development in weekly sprints, with each milestone delivering a functional increment. Several technical challenges required additional work:

Location validation. I implemented post-processing logic to cross-reference extracted values against known valid combinations.
Name normalization. Names appeared in various formats. I built a matching system that normalized these to canonical entries.
Historical name mapping. Records referenced entities that no longer exist under those names. I created a mapping system that translated historical names to current equivalents while preserving the historical record.

Technical Approach

Stack

React: Component-based architecture made the UI easy to extend as requirements evolved.
Golang: Type-safe, fast, and straightforward to maintain.
TypeScript: Type safety reduces bugs and makes collaboration easier.
Python: The original proof of concept was built in Python, so I reused as much of that code as possible for speed.
AWS: Managed services for hosting, image processing, and OCR reduced infrastructure overhead.
Docker: Each service ran in its own container, making the system modular.
GitHub Actions: Automated builds and deployments to catch issues early.
Keycloak: Handled authentication without building it from scratch.

Post-MVP

Analytics

I implemented usage analytics to provide visibility into processing performance. The client used this data for their own use and to report to their board on processing time per document, correction rates, and staff adoption.

External Integrations

In addition to CSV export, I integrated the application directly with an industry standard data repository. This integration included the processed data and original images, eliminating a manual step from the client's workflow.

Partnership Interest

Following deployment, the client received inquiries from peer organizations in their industry. Other firms requested access to the same processing capabilities. Software vendors in adjacent spaces expressed interest in integration partnerships.

Results

Reduced document processing time from 3 minutes to under 10 seconds through AI-assisted extraction with human review.
Recovered a stalled project and delivered a production system within two months of proof of concept approval.
Positioned the client as an industry reference for AI-powered document processing, attracting partnership interest from competitors and vendors.

Conclusion

This project recovered a stalled compliance initiative through proof-of-concept validation followed by focused execution. The system reduced processing time by over 90% and transformed a cost center into a competitive differentiator.

Key outcomes:

AI-powered extraction reduced processing from 3 minutes to under 10 seconds
Delivered production system within two months of POC approval
Client positioned as industry innovation reference

For document-intensive compliance requirements, this demonstrates feasibility validation before scale investment and the potential for regulatory requirements to become competitive advantages.