Resources
Get ideas and inspiration here!

  • Browse our AgentX Idea List for sample projects showing how to:

    • Index images as vectors, full-text index PDFs, and query them with a single API call

    • Chain multiple LLM calls for retrieval, summarization, decision logic, and action

    • Use MCP (Model Context Protocol) to distribute workloads across regions or clusters

    • Integrate external services, for example, payment gateways, mapping APIs, email auto-senders

  • Explore these sample applications built with TiDB Serverless:

    • Smart Service Desk (vector + full-text + LLM)

    • AI Study Buddy (syllabus ingestion + quiz generation)

    • Inventory Prediction Agent (vector clustering + LLM-generated restock orders)

Example Project Ideas

  1. Field Service Agent: Applications that ingest an image of a broken part (or a maintenance ticket), use vector search to surface similar repair manuals or past tickets, run a full-text query to find precise troubleshooting steps, and then chain a LLM call to guide a technician through each diagnostic action.

  2. Study Companion: Applications that take a course syllabus (vectorize each section) and textbook excerpts (full-text index), answer student questions by running vector + full-text searches for context, then call a LLM to generate practice quizzes, hints, or next-reading suggestions. If the student needs to solve a math problem, the agent calls an external calculator API.

  3. Supply Chain Orchestrator: Applications that index sales orders and inventory data in TiDB (vectors for demand patterns, full-text for product details), run vector search to predict reorder points based on historical trends, use a LLM to draft and send purchase orders automatically, and invoke an external workflow API to schedule follow-up tasks.

  4. Customer Support Workflow: Applications that process incoming support tickets (vectorize ticket text), use vector search to find similar past tickets, run full-text search to extract relevant knowledge-base passages, and then call a LLM to compose automated, step-by-step email responses or recommended actions.

  5. Knowledge Orchestration: Applications that ingest a variety of data sources (PDFs, CSVs, documentation) into TiDB’s vector and full-text indexes, allow a user to ask free-form questions, then combine vector + full-text results to craft detailed summaries or reports via LLM calls. These agents might also call external APIs (e.g., news feeds, charting services) to enrich the final output.

  6. Domain Specific Copilots: Applications built for a specific vertical (legal research, medical notes, finance dashboards) that use TiDB’s search capabilities to retrieve domain documents, then chain multiple LLM calls to analyze, summarize, or generate actionable guidance. These copilots should layer on at least one external API or multi-step logic (for example, fetching a regulatory code, analyzing it, and drafting a compliance checklist).

Note: These categories are meant as inspiration only. We encourage you to build any multi-step, agentic workflow that combines TiDB’s vector/full-text search, Model Context Protocol, and LLM calls to solve a real-world problem.

We have a number of resources to help you get started building your application:

TOOLS AND TECHNOLOGIES

  • Kimi Free 200 Million Token Resource:

    • Kimi, our Hackathon sponsor, is offering 200 million free tokens to every participating team. Note: Each team can only claim the token resource once.

    • How to Apply:

      • Join our Discord  
      • Go to the #sponsor-kimi channel  
      • Post a message following the required format to apply  
      • Once your application is approved, we’ll send your redeem code to you via Discord DM. Follow the guide below to claim your resources.
    • How to Claim:  
      • Log in to the Kimi Developer Platform → Go to Voucher Management → Select Redeem Voucher
      • Enter your redeem code and click Redeem