Skip to content

assignment.md

Latest commit

 

History

History
52 lines (32 loc) · 1.49 KB

assignment.md

File metadata and controls

52 lines (32 loc) · 1.49 KB

Assignment — Backend: Locate & Extract Tables/Images

Input

  • mmd_lines_data.json — layout metadata (pages, blocks, coordinates).
  • manual.mmd — Markdown manual (tables preserved, images referenced).

Task

Build a backend service that can:

  1. Locate the relevant table or image for a natural operator-style request (e.g., “Can you show me the drawing of a typical control valve setup?”).

  2. Return the table/image content.

  3. Include citation data with at least:

    • page_no
    • bounding box coordinates

Handle ambiguity (multiple matches) by returning top candidates; respond with "insufficient_info" if no confident match.

Example Operator Queries

  • “Can you pull up the comparison of actuator types?”
  • “Show me the sizing factors for liquids.”
  • “I need the noise level reference values.”
  • “Do we have a figure that explains cavitation?”
  • “Show me the overall oil & gas value chain diagram.”

Deliverables

  • API implementation.
  • DESIGN.md (≤1 page).
  • Sample outputs for 3–4 queries.

Note on Implementation

We prefer candidates who can effectively leverage coding agents (e.g. Cursor, Claude Code, or similar AI-assisted IDEs) to produce modular, well-structured, and high-quality code.

  • This means clean separation of concerns (indexing, retrieval, response formatting).
  • Meaningful function/module boundaries.
  • Readable, maintainable code, not just a script that “works once.”