SELECTED WORK · GLOBAL SAAS

Automation measured at feature level, not by demo impression.

Rule-based expert systems and geospatial automation supported more than 2,000 monthly map updates for The Home Depot at approximately 99% feature-level accuracy and 80% less manual adjustment.

Previous professional experience · Pointr

SETTINGGlobal indoor-location SaaS
SCALE2,000+ updates per month
QUALITY≈99% feature-level accuracy
IMPACT80% less manual adjustment

THE PROBLEM

Production scale exposes every ambiguous rule and inconsistent label.

Indoor-map content had to be classified, digitised and corrected across repeated batches. Manual work limited scale, while a single aggregate score could hide which feature types failed and why.

The goal was not maximum automation. It was controlled automation with measurable correction.

MY CONTRIBUTION

Combine rules, geospatial reasoning and quality evidence.

  • Designed and developed rule-based expert-system and geospatial automation in a cloud-based service architecture.
  • Created classification and digitisation logic for indoor-map features.
  • Built analysis distinguishing automated and manual changes across production batches.
  • Directed spatial data preparation, labelling, correction and quality control.
  • Authored a detailed labelling guide and coordinated AI benchmarking.
  • Contributed to microservice transition, R&D projects and patents.

THE ENGINEERING APPROACH

Quality had to be observable across the production loop.

01

Combine domain rules with data-driven methods

Use explicit spatial and business rules where they improve control, and model-based methods where data supports them.

02

Measure at feature level

Evaluate feature types and failure patterns instead of relying on one headline score.

03

Analyse change across batches

Separate what automation produced from what people corrected to expose recurring weaknesses.

04

Treat labelling guidance as architecture

Make domain meaning and annotation decisions repeatable across people and time.

OUTCOME

High-volume production with a measurable human correction boundary.

The automation supported more than 2,000 monthly updates for The Home Depot, reached approximately 99% feature-level accuracy and reduced manual map adjustment by 80%.

WHY THIS MATTERS ELSEWHERE

AI systems need correction evidence, not just output scores.

The same loop applies to document extraction, computer vision, geospatial processing, content operations and human-in-the-loop AI: measure what automation changes, what experts reverse and which rules or labels caused the difference.

Relevant capabilities: expert systems · geospatial automation · data quality · labelling design · AI benchmarking · production analytics · cloud services.

Public evidence note. The Home Depot is the only customer named in this public version. Other customer and internal implementation details are intentionally omitted.

RELATED DECISION

Can your automation explain where human correction remains necessary?