> ## Documentation Index > Fetch the complete documentation index at: https://docs.rotastellar.com/llms.txt > Use this file to discover all available pages before exploring further. # Workload Presets > 5 ready-to-use orbital compute workloads # Workload Presets Presets are complete workload definitions you can use immediately with `POST /v1/plan`. Each defines a multi-step pipeline with resource requirements, dependencies, security policies, and optimization objectives. List all presets: ```bash theme={null} curl https://rotastellar-cae.subhadip-mitra.workers.dev/v1/presets \ -H "Origin: https://rotastellar.com" ``` ## On-Board ML Inference All computation on-board. Captures 2GB of sensor data, runs ML inference, and downlinks only the 10.5MB encrypted result — a **190:1 data reduction**. | Property | Value | | ------------ | ----------------------------------------------- | | ID | `onboard-ml-inference` | | Category | ml-inference | | Steps | 4 (capture, preprocess, inference, encrypt) | | All on-board | Yes | | Data flow | 2,000 MB → 10.5 MB | | Policy | `min_latency`, 3-orbit deadline, 99% confidence | **Pipeline:** Sensor Capture → Data Preprocessing & Calibration → ML Model Inference → Encrypt Results ```bash theme={null} curl -X POST https://rotastellar-cae.subhadip-mitra.workers.dev/v1/plan \ -H "Content-Type: application/json" \ -H "Origin: https://rotastellar.com" \ -d '{"satellite_id": "25544", "preset_id": "onboard-ml-inference"}' ``` *** ## Split Learning Pipeline Bidirectional training. Satellite runs the first 3 neural network layers (feature extraction, 40:1 reduction), downlinks 36.75MB of activations. Ground trains the remaining layers and uplinks 5.25MB of updated weights. | Property | Value | | -------- | ----------------------------------------------------------------------------------------------------------- | | ID | `split-learning` | | Category | ml-training | | Steps | 9 (capture → feature extraction → compress → encrypt → train backend → compress weights → encrypt → deploy) | | Downlink | 36.75 MB (activations) | | Uplink | 5.25 MB (weights) | | Policy | `balanced`, 6-orbit deadline, 95% confidence | | Security | `confidential`, authenticated uplink, key rotation every 12 orbits | ```bash theme={null} curl -X POST https://rotastellar-cae.subhadip-mitra.workers.dev/v1/plan \ -H "Content-Type: application/json" \ -H "Origin: https://rotastellar.com" \ -d '{"satellite_id": "25544", "preset_id": "split-learning"}' ``` *** ## Earth Observation with QA Captures 5GB of imagery, runs on-board quality assurance to discard bad frames and cloudy scenes, compresses to 400MB, applies Reed-Solomon FEC and AES-256, then downlinks 560MB across multiple ground station passes. | Property | Value | | ------------------- | -------------------------------------------------------------------------------------------------------- | | ID | `earth-observation-qa` | | Category | earth-observation | | Steps | 8 (capture → QA → cloud filter → JPEG2000 compress → FEC encode → encrypt → ground validation → archive) | | Data flow | 5,000 MB → 560 MB transferred | | Multi-pass downlink | Yes | | Policy | `max_reliability`, 8-orbit deadline, 95% confidence | ```bash theme={null} curl -X POST https://rotastellar-cae.subhadip-mitra.workers.dev/v1/plan \ -H "Content-Type: application/json" \ -H "Origin: https://rotastellar.com" \ -d '{"satellite_id": "25544", "preset_id": "earth-observation-qa"}' ``` *** ## Federated Learning Privacy-preserving distributed training. The satellite trains locally on 500MB of data, computes and sparsifies gradients (top-k, 90% zeros), downlinks 3.7MB. Ground aggregates via FedAvg and uplinks 5.8MB updated global model. **Raw data never leaves the satellite.** | Property | Value | | -------- | ------------------------------------------------------------------------------------------------------------ | | ID | `federated-learning` | | Category | ml-training | | Steps | 10 (local train → gradients → sparsify → compress → encrypt → aggregate → compress model → encrypt → deploy) | | Downlink | 3.7 MB (sparse gradients) | | Uplink | 5.8 MB (global model) | | Policy | `balanced`, 6-orbit deadline, 95% confidence | | Security | `confidential`, authenticated uplink, key rotation every 12 orbits | ```bash theme={null} curl -X POST https://rotastellar-cae.subhadip-mitra.workers.dev/v1/plan \ -H "Content-Type: application/json" \ -H "Origin: https://rotastellar.com" \ -d '{"satellite_id": "25544", "preset_id": "federated-learning"}' ``` *** ## Resilient Store-and-Forward Relay Receives 100MB from a remote sensor during one pass, applies Reed-Solomon erasure coding (rate 2/3 — any 2-of-3 blocks reconstruct), buffers on-board, and transmits during a different ground pass. | Property | Value | | ---------------- | ---------------------------------------------------------------------------------------- | | ID | `resilient-store-forward` | | Category | relay | | Steps | 5 (uplink receive → integrity check → erasure coding → encrypt & buffer → ground decode) | | Data transferred | 157.5 MB (with erasure coding overhead) | | Policy | `max_reliability`, 4-orbit deadline, 99% confidence | ```bash theme={null} curl -X POST https://rotastellar-cae.subhadip-mitra.workers.dev/v1/plan \ -H "Content-Type: application/json" \ -H "Origin: https://rotastellar.com" \ -d '{"satellite_id": "25544", "preset_id": "resilient-store-forward"}' ``` *** ## Comparison | Preset | Steps | Data Reduction | Downlink | Uplink | Objective | Deadline | | --------------------- | ----- | ----------------- | -------- | ------- | ---------------- | -------- | | On-Board ML Inference | 4 | 190:1 | 10.5 MB | — | min\_latency | 3 orbits | | Split Learning | 9 | 40:1 | 36.75 MB | 5.25 MB | balanced | 6 orbits | | Earth Observation QA | 8 | 9:1 | 560 MB | — | max\_reliability | 8 orbits | | Federated Learning | 10 | 135:1 | 3.7 MB | 5.8 MB | balanced | 6 orbits | | Store-and-Forward | 5 | 1:1.6 (expansion) | 157.5 MB | — | max\_reliability | 4 orbits | Need something different? Define a [custom workload](/cae/custom-workloads).