Case study · Banking · AI
Utkarsh Small Finance Bank Masked 2.8 Lakh Aadhaar Documents in One Week
Utkarsh Small Finance Bank is an RBI-licensed small finance bank serving a large and growing customer base across the country. RBI data masking regulations applied to the over 2.8 lakh Aadhaar documents stored in its document management system (DMS).
Before masking could begin at scale, the bank needed to know which documents had already been masked and which still needed processing. Then the full masking workflow had to run accurately, securely and within the compliance window.
- 99% masking accuracy
- 2.8 lakh documents in one week
- No data left the bank
Results at a glance
99%
Masking accuracy across the full repository
2.8 lakh
Aadhaar documents masked to RBI standards
1 week
To complete the masking, without disrupting banking operations
The Challenge
The bank's path to RBI compliance involved several connected challenges, each of which shaped what a workable solution had to do.
01
Identifying Pre-Masked Files
The repository held files already partly or fully masked by earlier processes alongside files that still needed masking. Without a reliable way to tell them apart, the bank risked either reprocessing compliant documents or leaving gaps in coverage. An accurate inventory of masking status across all 2.8 lakh files had to come before any masking.
02
Offline Data Security and Compliance
The bank's security policy was clear: no data could leave the building. Processing had to run in a completely offline environment with no internet access at any point. That ruled out cloud-based or externally hosted options and constrained the architecture from the start.
03
Limited Training Data for the AI Model
Accurate Aadhaar detection and masking models usually need substantial training data. Utkarsh could provide 200 sample files, so model development needed a more resourceful approach to reach the accuracy regulatory compliance required.
04
Rapid Processing of 2.8 Lakh+ Documents
The document volume and the strict RBI timeline called for processing built for speed. It also had to run without disrupting core banking operations: throughput had to be high, and the process had to run reliably from start to finish.
The Solution
HabileLabs built an end-to-end AI masking pipeline around Utkarsh's specific constraints: offline operation, limited training data, a large document volume, and a defined compliance deadline.
Intelligent Identification of Pre-Masked Files
HabileLabs used AI and OCR to analyze the full repository sequentially and report which files were already masked. Only documents that genuinely needed processing entered the masking pipeline, which avoided redundant work and improved efficiency.
AI Model Development with Limited Data
Using data augmentation, HabileLabs built accurate models for Aadhaar number detection and QR code identification from the 200 sample files available. Augmentation expanded the effective training dataset, so the models performed reliably despite the limited starting point.
End-to-End Compliance Management
The solution covered the full compliance workflow, from identifying pre-masked files to applying the required masking. Audit logging and detailed reporting were part of the delivery, giving the bank a documented record of processing outcomes across the repository.
Secure Offline Deployment with Docker
The masking pipeline was containerized with Docker and deployed within the bank's internal infrastructure. No data left the bank's offline environment at any stage, and the architecture met the bank's security requirements in full.
Parallel Processing on a 16-Core Server
To process 2.8 lakh documents within the compliance window, HabileLabs ran the pipeline on a 16-core server configured for parallel processing. That throughput completed the masking in one week without disrupting live banking operations.
Business Impact
99% Masking Accuracy
The solution delivered the accuracy RBI compliance requires across the full 2.8 lakh document repository, significantly reducing the risk of regulatory penalties.
2.8 Lakh Documents Processed in One Week
Parallel processing completed the full masking operation within the compliance window, with no disruption to core banking operations.
Reduced Operational Costs and Resource Use
Identifying pre-masked files upfront avoided redundant processing across part of the repository, saving computing resources and time.
Stronger Data Security and Customer Trust
Sensitive Aadhaar data for over 2.8 lakh customers was processed entirely within the bank's offline environment and never left its infrastructure.
Compliance and Risk Mitigation
Full masking coverage at 99% accuracy addressed the bank's RBI compliance obligations and reduced its exposure to regulatory risk.
Audit-Ready Reporting Across 2.8 Lakh Documents
Detailed processing reports gave the bank's compliance and governance teams a clear, documented view of masking outcomes across the repository.
Why Utkarsh Small Finance Bank Chose HabileLabs
Proven AI Capability Under Real-World Data Constraints
In a compliance project like this, model accuracy is non-negotiable, and most vendors need thousands of labeled samples to get there. Utkarsh needed a partner that could work with what was available: 200 sample files. HabileLabs' use of data augmentation to build compliance-grade models from that constrained dataset was the deciding technical factor.
Fluency in Indian Banking Regulation
Utkarsh needed a partner that already understood RBI data protection requirements and the security standards expected of regulated financial institutions, not one that would learn them during delivery. HabileLabs' knowledge of Indian banking regulation meant the solution was designed for compliance from the first conversation.
A Scope That Matched the Actual Problem
The requirement went beyond masking: it began with knowing what had already been masked. Utkarsh chose HabileLabs because the engagement covered the complete workflow, from pre-mask identification and active masking to audit logging and reporting, leaving nothing for the bank's team to stitch together.
Conclusion
Planning something similar?
Utkarsh Small Finance Bank came to HabileLabs with a clear mandate: mask 2.8 lakh Aadhaar documents to RBI standards, within a defined compliance window, in a completely offline environment. The constraints were real: limited training data, strict security requirements, and a large repository that had to be processed without disrupting banking operations.
The outcome was 99% accurate Aadhaar masking completed in one week, with full audit reporting, no data leaving the bank's infrastructure, and core banking functions unaffected throughout.