logo

Research & Evaluation

StarlineAI is designed to automate regulatory compliance checks for Indian architectural drawings. Given the fragmented, ambiguous, and jurisdiction-specific nature of building regulations in India, meaningful evaluation requires real-world data, expert review, and context-aware benchmarks.

Research and Evaluation

Dataset Description

  • 400+ real-world Indian architectural drawings spanning interiors, residential, mixed-use, and large-scale developments.
  • Drawings reflect actual submission standards, drafting practices, and jurisdictional constraints.

Relative depth of regulatory rule coverage across major compliance domains.

National Building Code
State-level regulations
Municipal by-laws
Zoning and land-use rules

Evaluation Methodology

Performance was assessed using a human-in-the-loop evaluation framework, where experienced architects manually reviewed drawings and compared their findings against StarlineAI outputs. Each detected issue was classified as a correct detection, false positive, or missed issue, with human review serving as the reference baseline reflecting real professional compliance workflows.

Cumulative progression of identified compliance issues through the automated detection and human review pipeline.

Performance Metrics

In comparative evaluations, a human reviewer typically identified approximately 80 compliance issues per drawing. StarlineAI identified 75 to 76 issues on average, of which 71 to 72 were confirmed as correct detections after filtering false positives.

Detection Quality

Recall approximately 94 percent, precision between 93 and 95 percent, with a false negative rate of 6 to 7 percent.

Time to Compliance

Automated checks reduce basic compliance review time from approximately 10 minutes to 3–4 minutes per drawing.

Overall Accuracy

90–92%

Observed accuracy in controlled evaluations. Results vary by drawing quality, jurisdiction, and rule complexity.

Failure Modes and Limitations

While StarlineAI demonstrates strong performance, certain regulatory and contextual limitations remain inherent to automated compliance analysis.

Ambiguous or subjective code language
City-specific regulatory edge cases
Dependence on drawing quality and annotation standards
Conservative flagging for borderline issues
Not a statutory municipal approval authority

Benchmarking Context

There is currently no publicly available global benchmark for automated building code compliance systems. StarlineAI’s evaluation approach emphasizes transparency, real-world data, and India-specific regulatory contexts.