clarify message

Inspiration

This project started with a personal frustration: missing important deadlines because a short notice was buried in a busy group chat or posted in an app I rarely opened. Freelancers and small teams face a similar problem when client requests, schedule changes, and follow-ups are scattered across conversations.

What it does

Clarify Message helps users turn work messages into a reviewable action list. Users paste messages or import a text file. The application can identify action items, suggest priorities, and show the original messages behind each recommendation.

In AI mode, it uses DeepSeek to analyze context, recognize explicit deadline changes, and identify situations where someone is waiting for a response. Users review and edit the results before confirming tasks, setting reminders, or exporting them to a calendar.

The application also includes local keyword screening and a clearly labeled, preconfigured demonstration that works without an API key.

How we built it

The application uses React and TypeScript, with Vinext and a Cloudflare Workers-compatible backend. The AI endpoint validates requests, calls DeepSeek, and checks that quoted evidence exists in the original input.

Messages and tasks are saved in the current browser. API keys remain in page memory and are not saved in browser storage. Calendar export uses the standard ICS format.

Challenges we ran into

The main challenge was avoiding false certainty. Messages can contain vague dates, revised deadlines, or requests intended for someone else. The application keeps uncertain times unresolved, preserves source evidence, and requires human confirmation.

Another challenge was reminder reliability. In-page reminders require the page to remain running, so users can also export confirmed tasks to their calendar.

Accomplishments that we're proud of

We built a working flow from message input to task review, confirmation, reminders, and completion. Core checks and desktop/mobile browser tests passed, covering evidence validation, invalid dates, calendar export, persistence, and reminder state.

Live paid AI calls still require validation with a valid user-provided DeepSeek API key. The preconfigured demonstration is explicitly labeled and is not presented as a live AI result.

What we learned

Useful AI assistance needs traceable evidence, clear uncertainty, and user control. A recommendation becomes more actionable when users can see why it was made and correct it before relying on it.

What's next for clarify message

We plan to evaluate the AI analysis on consented, anonymized work conversations and investigate supported ways to receive messages automatically. The current version uses manual input and does not automatically read WeChat, QQ, or other apps.

Built With

• data.gov.sg
• efficiency
• helper