| I&T Solution |
AI-assisted Data Submission Analysis System for MEELS
(REF: S-2106) |
| Trial Project |
|
| Solution Feature |
- The proposed system combines document intelligence, workflow automation, and human-in-the-loop verification to automate form entry, report interpretation, certificate checking, signature/chop validation, and exception triage. It reduces repetitive vetting effort, shortens turnaround time, improves consistency, and retains officer control through measurable digital workflow controls.
- A modular AI pipeline supports printed text, bilingual handwriting, tables, scanned certificates, company chops, signatures, and supplementary files. It produces structured fields, confidence scores, page coordinates, and evidence links, so reviewers can trace each extracted value back to the original document region and confirm it quickly.
- Built-in cross-checking connects extracted data with existing departmental records and certificate validity sources. Rules compare model numbers, applicant identities, test parameters, dates, certificate references, and consistency across pages, with higher thresholds for critical fields to reduce manual rework and improve regulatory confidence during routine processing.
- The web-based workbench provides case queues, side-by-side evidence viewing, editable field grids, discrepancy explanations, report drafting, applicant prompt generation, dashboards, and full audit logs. It supports daily vetting operations, supervisory review, and management-level monitoring of workload, accuracy, ageing cases, and exception trends without switching screens.
- To address limited labelled historical data, the system applies a semi-supervised joint stochastic approximation (JSA) learning loop. Confirmed corrections provide labels, while unlabelled submissions contribute latent evidence signals, enabling lower-variance model updates, better retrieval-extraction alignment, and continuous accuracy gains without large manually labelled datasets.
|
| Trial Application and Expected Outcome |
- The nine-month trial starts with requirement confirmation, sample collection, environment preparation, and data-security alignment. A representative document set will be agreed, including application forms, laboratory reports, business registration certificates, handwritten forms, scanned copies, and supplementary evidence, so testing reflects realistic operational conditions and acceptance scenarios.
- Pilot implementation will first cover ingestion, preprocessing, OCR, table parsing, chop/signature detection, field extraction, database comparison, and officer review. Accuracy baselines will be measured by document type, scan quality, printed text, handwriting, and critical parameters, then refined through controlled review feedback and exception analysis.
- User acceptance will be conducted through real cases, shadow-mode comparison, officer correction logs, and management dashboards. Human reviewers remain the decision authority during the pilot, while the system records evidence, confidence, exceptions, processing time, and improvement opportunities for measurable trial evaluation and final reporting.
- R&D outcomes will be realized through configurable schemas, rule catalogues, model registry, annotation repository, and deployment playbooks. These reusable assets allow the pilot system to become a maintainable production-ready platform after validation, with future document types and policy changes handled through configuration and governance.
- Final delivery will include trained models, web workflow, dashboard, integration interfaces, self learning procedures, test reports, user manuals, administration guide, and transition plan. Accuracy, 2 performance, usability, security, and operational readiness results will be summarized to support production rollout and long-term maintenance planning after trial completion.
|
| Info on I&T Solution Provider |
| Solution Provider | : | Hong Kong Tasitech International Limited | | Address | : | Unit 630, 6/F., Building 16W Hong Kong Science Park, Shatin, New Territories, Hong Kong | | Contact Person | : | Victor Sze |
| Position | : | Business Development Director | | Tel | : | 67922983 | | Email | : |
victorsze@tasitech.ai | | Webpage | : | www.tasitech.com.cn |
|
