OurTasker

Project: OurTasker ✨ About the Project

What inspired us We were inspired by the challenge of going beyond simple chatbots or one-step RAG demos. In real life, people don’t just want answers — they want actions. That’s why we built OurTasker, an AI agent that feels like “our shared assistant,” turning raw data into meaningful workflows.

What we learned

How to combine TiDB Serverless vector search with full-text search for richer results.

How to chain LLM reasoning with external APIs to make agents not just informative, but useful.

The importance of keeping workflows simple, so users can trust and adopt them. How we built it

Ingested and indexed data (documents, logs, or messages) into TiDB Cloud.

Queried the data with vector + text search to find relevant matches.

Chained LLM calls to analyze the results, summarize findings, and plan actions.

Invoked external APIs (e.g., Slack, Google Sheets, or Twilio) so the agent could complete a task from start to finish.

Challenges we faced

Designing multi-step flows that are reliable and don’t break mid-process.

Handling noisy or incomplete data and still producing useful results.

Balancing speed and accuracy in workflows without overwhelming the user.

Built With

• google-sheets-api
• javascript-(node.js)-frameworks:-fastapi-(backend)
• languages:-python
• react-+-tailwind-(frontend)-databases:-tidb-serverless-(vector-+-full-text-search)-llms:-openai-gpt-4o-(or-claude-/-llama-3.1)-cloud-services:-tidb-cloud
• render/aws-(backend)-apis-&-tools:-slack-api
• twilio
• vercel/netlify-(frontend-hosting)