Inspiration

Industrial manufacturing systems generate rich operational data, but most operator experiences are fragmented across static dashboards, technical logs, and siloed tools. We were inspired to design a single spatial interface where machine understanding feels immediate, intuitive, and elegant. The vision became a digital co-pilot for the HP Metal Jet S100: blend real-time telemetry, predictive insight, and contextual AI into one “always-on” control surface.

What it does

Digital Co-Pilot provides a unified digital twin experience with:

  • A 2D schematic and 3D digital twin view that switch seamlessly
  • Live component monitoring for key subsystems (build unit, printhead, recoater, etc.)
  • Contextual component focus cards with status, alerts, and metrics
  • A summonable Aether chat that can answer general or part-specific questions
  • Voice input via speech-to-text for hands-free operation
  • Smooth camera focus and progressive disclosure to inspect internals safely and clearly

How we built it

  • Backend: FastAPI service exposing twin, telemetry, and agent endpoints
  • Simulation/Data layer: Synthetic degradation simulator + long-horizon validation trajectories
  • AI layer: Grounded diagnostic agent with telemetry-aware responses and evidence-linked reasoning
  • Frontend: React + TypeScript dashboard with:
  • 3D scene (React Three Fiber / three.js)
  • 2D schematic mode
  • Analytics tiles (trajectory, ranking, alerts)
  • Stateful interaction model (selection, focus, time controls, chat context)
  • Voice: Speech-to-text and text-to-speech integrations for operator UX
  • State + UX: Zustand store, motion transitions, progressive disclosure, and contextual overlays

Challenges we ran into

  • Rendering stability in 3D transitions (black-screen/context lifecycle issues)
  • Keeping 2D, 3D, and analytics views perfectly synchronized to one source of truth
  • Designing camera behavior that is inspectable but not disorienting
  • Communicating forecast semantics clearly (e.g., maintenance/reset effects vs “flatline” -misconceptions)
  • Balancing rich intelligence output with concise, operator-friendly explanations
  • Integrating proactive alerts, agent context, and temporal simulation into one coherent flow

Accomplishments that we're proud of

  • Built a polished digital twin UX that feels like one product, not a set of tools
  • Delivered both required dimensions:
  • Functionality: model + simulate + interact
  • Intelligence: predict + explain + recommend + verify
  • Achieved smooth cross-view interaction and contextual component drill-down
  • Added grounded AI assistance that remains useful during multi-turn troubleshooting
  • Created a demo flow that clearly shows operational value, not just technicalities

What we learned

  • In industrial UX, clarity beats feature count: context and prioritization are everything
  • Predictive systems must explain why, not just output risk scores
  • Temporal simulation is powerful only when tied to concrete action loops
  • Spatial interfaces (2D/3D) increase trust when interactions are deterministic and consistent
  • “AI co-pilot” value comes from grounding + actionability, not conversational flair

What's next for Digital Co-Pilot for HP Metal Jet S100

  • Add scenario comparison (maintenance policy A vs B over future horizons)
  • Improve explainability with causal traces and confidence decomposition
  • Expand proactive workflows (auto-generated maintenance work orders)
  • Add role-based experiences (operator, reliability engineer, plant manager)
  • Integrate with real historian/MES streams for production pilots
  • Support multi-printer fleet orchestration and plant-level optimization dashboards
Share this project:

Updates