The ₹300-Crore Error Trap: NHAI Deploys In-House AI to Catch Costly Highway Planning Gaps Before Ground Is Broken
A new Technical Schedule Analyzer and the AI assistant Margsarthi are shifting the authority’s approach from reactive dispute resolution to proactive plan verification, targeting discrepancies in detailed project reports that have historically driven major cost overruns.
NEW DELHI: The National Highways Authority of India (NHAI) is embedding artificial intelligence directly into the pre-construction workflow in a bid to eliminate errors that routinely add hundreds of crores to project costs. At the center of the effort is the Technical Schedule Analyzer, an in-house software tool designed to scrutinize detailed project reports (DPRs) by comparing quantities and specifications against Indian Roads Congress codes, Ministry of Road Transport and Highways norms, and NHAI’s own standards before any earth is moved.
The financial stakes are material. NHAI chairman Santosh Kumar Yadav told the Hindustan Times that variations of ₹200 crore to ₹300 crore can surface in a ₹2,000-crore project simply because of omissions during the planning phase. The analyzer, which currently examines Schedule B and Schedule C sections of DPRs, flags inconsistencies and missing elements at the blueprint stage, compressing a review cycle that previously relied on manual cross-referencing of hundreds of pages of technical documents. The next phase will integrate drone survey data, Network Survey Vehicle records, geospatial information, monthly progress reports, and site photographs, allowing the system to verify whether what was planned on paper matches the actual conditions along an alignment.
The analyzer is one application within a broader AI ecosystem being built by NHAI’s in-house AI Cell, established in August 2025 after an internal review found that project directors and regional officers were spending disproportionate time on administrative disputes and document interpretation instead of site supervision. A multidisciplinary team of civil engineers, AI specialists, and designers digitized thousands of circulars, manuals, and policy documents into a structured knowledge repository. That repository now powers Margsarthi, a closed-environment AI assistant launched in April that has already handled more than 50,000 queries from approximately 1,100 users across the organization. Every response cites the source document, letting engineers verify the basis of any recommendation.
A key use case has been resolving technical disagreements at project sites. When contractors and authority engineers interpret a provision differently, both parties can query Margsarthi, which points them to the relevant IRC code or policy clause, collapsing hours of manual search and debate into minutes. The platform is also being used to analyze lengthy legal and arbitration records, and a recent photo-based diagnostic feature allowed an engineer to upload an image of an improperly cut slope and receive corrective guidance tied to applicable standards. Other tools under development include automated road defect detection, maintenance scheduling, AI-assisted drawing reviews, and support for environmental and forest clearance applications.
Officials stressed that the systems are designed to inform decisions, not replace the judgment of engineers. Accountability for project outcomes remains with the officers, even as AI-generated insights become embedded in daily operations.
BuiltWorld AI Operational Take: NHAI’s approach highlights a practical path for public infrastructure agencies where the immediate AI payoff lies not in autonomous design but in dramatically reducing the information retrieval and compliance-checking friction that breeds contractual disputes. The closed, source-verified architecture of Margsarthi addresses the hallucination risk that makes generic AI tools unusable in regulatory contexts. The long-term test will be whether field data integration can shift the system from catching errors in documents to predicting them before the DPR is finalized, moving the entire project lifecycle from forensic correction to anticipatory governance.