Engineering Diagram Graph Annotator

About OpenTrain

OpenTrain is the #1 platform for building careers in AI training and data labeling. We connect skilled annotators with projects that teach and shape modern AI systems, and we manage contracts, guidelines, and quality checks so you can focus on accurate labeling.

Why AI training matters

AI training (data labeling/annotation) is the human work that makes machine learning systems reliable: people create the examples and structure models learn from. Contributors work fully remotely, often with flexible hours, and directly influence how state-of-the-art models behave.

The role

You will convert engineering diagram sheets into graph-style annotations: identify every symbol as a node (with a classified bounding box) and mark connectivity between nodes as edges. The job requires careful, pixel-accurate boxes on large images and labeling relationships (region-to-region connections).

• Node classes: valve, pump, instrumentation, general, tank, arrow, inlet/outlet.
• Routing/helper nodes: connector (line bends/junctions) and crossing (line crossings).
• Edges: label each edge representing connectivity as solid or non-solid.
• Boxes must be drawn at full resolution and aligned to the supplied image coordinate frame.

Deliverables & volume

Deliver clean, consistent annotations exported in one of the following formats (final tool TBD): Label Studio JSON, CVAT XML, or COCO. Exports must use the same coordinate frame as the supplied images; OpenTrain will handle any downstream conversion.

This project is approximately 60 sheets in total. Work begins with a small paid pilot before the full batch is assigned.

• Data type: IMAGE. Label types: BOUNDING_BOX / OBJECT_DETECTION.
• Expected employment type: Contractor. Worldwide applicants accepted.
• Project payment: fixed price USD 499 (see pilot details).

Requirements (must-haves)

Candidates must demonstrate hands-on experience with annotation tools that support both bounding boxes and region-to-region relations, a strong ability to read technical/engineering diagrams, and a disciplined approach to consistent labeling and QC.

• Tooling proficiency: experience using Label Studio, CVAT, or an equivalent tool for boxes and relations.
• Diagram comprehension: comfortable reading engineering diagrams and distinguishing symbols from connecting lines.
• Classification consistency: follows a written taxonomy and flags ambiguous symbols rather than guessing.
• Bounding-box precision: able to place tight, pixel-accurate boxes on large, dense images.
• Connectivity labeling: experience labeling relationships/edges (not just objects).

Nice-to-have

Prior work on object-detection datasets or graph/relationship annotations is a plus and will help during screening and the pilot.

• Experience exporting Label Studio JSON, CVAT XML, or COCO for downstream use.
• Previous projects annotating diagrams, schematics, or graphs.

Pilot, QC, and throughput expectations

We require a paid 1-sheet pilot to verify tooling, label consistency, and connectivity annotation. QC feedback may require iteration; candidates should be responsive and able to apply corrections across the batch.

During screening you will be asked for a realistic throughput estimate and a plan to keep quality consistent across a multi-sheet batch (self-QC passes, break strategy, re-check of connectivity).

• Pass criteria: meets all must-haves (tooling, diagram comprehension, consistency, precision, connectivity) and gives acceptable answers on throughput and QC responsiveness.
• Auto-screen red flags: no relation/connection labeling experience, cannot name a supported tool, guesses ambiguous symbols instead of flagging, or gives unrealistic throughput with no QC process.

How to apply & screening questions

Apply through OpenTrain with a short profile and answers to the screening questions below. Include any relevant past annotation samples and which annotation tools you've used. Successful applicants will be invited to the paid 1-sheet pilot.

• Which annotation tools have you used, and on which ones have you drawn both bounding boxes and connections/relations between regions?
• When two symbols are joined by a line, how do you decide where each symbol's bounding box ends and the connecting edge begins?
• How do you handle a symbol that doesn't clearly fit any class in the provided legend?
• What's your approach to keeping boxes pixel-accurate on large, dense images?
• Describe a time you labeled relationships or connections between objects, not just the objects themselves.
• Roughly how many densely-labeled sheets can you complete per week at high quality, and how do you keep quality consistent across a batch?
• If a QC review flags that 10% of your boxes are too loose or some edges are missing, how would you respond?