ProductionAI Experiment2026
Rice Doctor AI
AI-assisted ASEAN rice disease scanning app for early detection.
ReactAI/MLPython
Read case study
All projects
PrototypeAI Experiment2025
Smart Document Assistant
A prototype exploring how AI can organize and summarize complex documents while keeping a reviewer in the loop.
Problem
Managing complex information manually can create repetitive work and make important details difficult to find.
Context
Public-facing teams often receive long documents in inconsistent formats, and answering a simple question can require reading dozens of pages.
My role
Sole developer — product framing, data model, API, and interface.
Solution
A structured workflow combining document parsing, AI-assisted summarization, human review, and searchable structured output.
Key features
- Upload and parse documents into structured sections
- AI-generated draft summaries with source references
- Reviewer queue with approve, edit, and reject actions
- Full-text and field-level search across approved records
Architecture overview
- 1
Document upload
- 2
Validation & parsing
- 3
AI summarization
- 4
Human review
- 5
Searchable output
AI usage
The AI produces draft summaries and suggested fields. Every record must be approved by a person before it becomes searchable.
Accessibility
Keyboard-operable review queue, labelled form controls, and status communicated with text as well as color.
Security
Documents are scoped per account, uploads are validated by type and size, and AI calls run server-side so keys never reach the browser.
Outcome
Demonstrates how AI can reduce repetitive reading work without removing human accountability.
Lessons learned
- Structured output is far more useful than free-form summary text.
- A visible review state builds more trust than a higher-accuracy model.
- Showing the source passage next to each suggestion cut review time noticeably.
Related projects
ConceptAutomation Tool2025
Service Request Triage Workflow
A concept for routing incoming public service requests using rules first and AI assistance second.
TypeScriptReactZodNode.js
Read case study