FactTrail：以證據為本追蹤截止日期與要求
FactTrail — Evidence-First Tracking for Deadlines & Requirements
FactTrail 是面向學生及求職機會者的本機 AI 原型，可追蹤公開網頁上的截止日期、資格要求及費用，並比較頁面內容變化。它使用 Ollama 的 llama3.2:1b 提取候選資訊，以 SQLite 儲存資料，並透過 SHA-256 偵測變更，讓用戶檢視前後文字及相關證據。重新檢查目前須手動進行，FactTrail 尚未提供排程監察或自動通知。
FactTrail — Evidence-First Tracking for Deadlines & Requirements
Tagline: Know what changed. Make your next move with confidence.
Project Description
FactTrail is a local-first AI prototype designed to help students and opportunity seekers track important information published on public program webpages, including application deadlines, eligibility requirements, fees, and other details.
Instead of repeatedly checking pages manually, users can save public sources, extract candidate facts using a local AI model, and compare webpage content across manual rechecks. FactTrail records detected content changes and provides previous and current text for review.
The Problem
Important deadlines, eligibility rules, and application requirements can be difficult to keep track of across multiple websites. Manually revisiting pages takes time, and changes can be overlooked.
Our Solution
FactTrail combines local AI extraction, a SQLite database, and page-content comparison into one dashboard. It helps users organize sources, review extracted information, and identify when saved webpage content changes.
Key Features
- Public source collection: Add a webpage URL and attempt to fetch its content.
-
Local AI extraction: Use Ollama with the
llama3.2:1bmodel to extract candidate deadlines, eligibility details, and fees. - Evidence for review: Present supporting text and extraction statuses so users can review the extracted facts.
- Local storage: Save source records and page snapshots in SQLite.
- Change detection: Compare normalized webpage content using SHA-256 hashing.
- Change history: Display previously saved and newly fetched text when a content change is detected.
- Manual rechecking: Recheck an individual saved source or batch-recheck saved sources.
Technology Stack
- Next.js and React
- TypeScript
- Tailwind CSS
- Ollama — local AI inference
- SQLite — local data persistence
- SHA-256 — webpage-content comparison
What Works Today
The dashboard, source-fetching API, local extraction API, SQLite persistence, snapshot recording, change-history API, individual rechecks, and batch recheck endpoint are implemented.
The project passes ESLint, TypeScript checks, and the Next.js production build.
What Makes FactTrail Different
FactTrail is designed to run its AI extraction locally rather than requiring a paid AI API. It focuses on keeping source information, candidate facts, and content-change evidence together so that users can inspect the evidence before making decisions.
Current Limitations and Future Work
This is a working prototype, not a fully automated monitoring service. Rechecks are currently manual; scheduled background monitoring and automatic alerts are not implemented. A webpage-content change does not necessarily mean that a deadline or eligibility rule changed. Extracted facts and detected differences must be reviewed by a person.
Planned improvements include scheduled rechecks, notifications, stronger fact-level change detection, and further security testing.
Source Code
GitHub repository: https://github.com/jaganbaditya16-note/facttrail
Demo
The current demo runs locally at http://localhost:3000 after setup. A publicly hosted live demo has not been deployed yet.
Cost
Designed for a ₹0-cost local prototype using a locally hosted AI model, with internet access required to fetch public webpages.
FactTrail — Less checking. More confidence in your next move.
來源：DEV Community · dev.to