Prompt library

Upload every CSV file in ./data/idle as training samples with label "idle", then do the same for ./data/wave and ./data/punch

Upload every .wav file in ./recordings as testing samples with label "noise"

Upload the image at ./capture.jpg as a training sample with label "person" to my Edge Impulse project

Read all CSV files in ./inbox, convert each to Edge Impulse JSON format with 3-axis IMU data at 62.5 Hz, and upload them as training samples with label "running"

Generate 30 synthetic accelerometer samples for each of the labels "idle", "walking", and "running" and upload them as training data. Use 3-axis (accX, accY, accZ) at 62.5 Hz for 2 seconds per sample, with randomised values intended to approximate sensor noise.

List all training samples in my Edge Impulse project grouped by label, with counts

Show me all samples in my project where the label is "unknown" or empty

List all testing samples and show the filename and label for each

Show me which labels have fewer than 20 training samples

Find any training samples whose filename contains "test" — these may have been uploaded to the wrong category

Show me the 10 most recently uploaded samples and their labels

Relabel all training samples whose filename starts with "idle_" to "idle"

Delete all training samples with label "unknown"

Delete all samples shorter than 1 second from the training set

Move all training samples with label "background" to the testing set

Find training samples whose label is "walking" but whose filename contains "run", and ask me before relabelling each one

Deduplicate the training set — list any samples with identical filenames and ask me which to keep

Move 20% of training samples for each label to the testing set, picking them at random, making sure each label is represented proportionally

Check the training/testing split for each label and tell me if any label has fewer than 15% of its samples in the testing set

Show me the class balance across my training set and flag any label with fewer than half the samples of the largest class

Start a training job in my Edge Impulse project and wait for it to finish, then print the accuracy and loss

Start a training job and give me the job URL so I can monitor it while it runs

Start training, poll until it finishes, and if it fails show me the last 30 lines of the job log

Check whether there is a training job currently running in my project and show its status

Show me all jobs in my project from the last 24 hours, with their status and duration

The last training job failed — fetch the job log and show me the error lines

List all failed jobs in my project in the last 7 days and show the error message from each log

Run model testing on my Edge Impulse project and show the confusion matrix and per-class F1 scores

Show me all samples that were misclassified in the last model test, with their predicted and actual labels

Which class has the lowest F1 score in my current test results? Show me the misclassified samples for that class.

Compare the test accuracy of the last three training runs and tell me which performed best

Show me the on-device performance estimates from the latest trained model — RAM, flash, and inference time

Start an EON Tuner run on my Edge Impulse project targeting the Arduino Nano 33 BLE Sense and wait for it to finish

Start an EON Tuner run optimising for the smallest model that still achieves at least 90% accuracy

List all EON Tuner runs for my project and show the top 5 results ranked by accuracy

List all EON Tuner results ranked by inference time, and show only the ones with accuracy above 85%

Take the best EON Tuner result and set it as the active impulse configuration

Export a C++ library from my Edge Impulse project and save it to ./build

Export an Arduino library and save it to ./build/arduino-library.zip

Export a TFLite int8 model and save it as ./build/model.tflite

Export an ONNX model and save it to ./build/model.onnx

Export a Linux AARCH64 deployment (.eim) for the Raspberry Pi and save it to ./build

Export a deployment for Arduino UNO Q (GPU) and save the .eim file to ./build

Export a WebAssembly deployment and save it to ./build/wasm

Retrain my Edge Impulse project and re-export the C++ library to ./build when training finishes

Check if the model has been retrained since the library in ./build was exported. If so, download the latest version.

Write an Arduino sketch for the Nano 33 BLE Sense that reads the onboard IMU at 62.5 Hz and runs inference every 2 seconds, printing the top label and confidence to Serial

Write an Arduino sketch that reads from a PDM microphone, fills the feature buffer, runs inference, and blinks the built-in LED when the label "yes" is detected with confidence above 0.8

Add error handling to my existing sketch so it prints a message to Serial if run_classifier() returns anything other than EI_IMPULSE_OK

Write a C++ main.cpp that reads sensor data from a CSV file, runs inference using the Edge Impulse C++ library, and prints each label with its score

Write a C++ application that captures frames from a USB webcam using OpenCV, runs the image classifier, and overlays the top label and confidence on the video feed

Generate a CMakeLists.txt that links my main.cpp against the Edge Impulse C++ library extracted at ./ei-sdk

Write a Python script that loads the .eim model at ./build/model.eim and runs inference on every image in ./test-images, printing the top label for each

Write a Python script using the Edge Impulse Linux Python SDK that captures audio from the default microphone and runs keyword spotting inference in a loop, printing results to the console

Export my Edge Impulse C++ library, then write a small Node.js Express server that loads the .eim model, accepts a POST request with raw sensor data as JSON, runs inference, and returns the label and confidence scores as JSON

Build a simple HTML + JavaScript dashboard that calls my inference server every second, displays the latest predicted label, and shows a bar chart of confidence scores for each class using Chart.js

Write a Python Flask app that accepts image uploads, runs them through the Edge Impulse image classifier .eim, and returns the results as a JSON response with label and confidence

Write a Python Flask app that ingests my computer's web camera feed, runs the feed through the Edge Impulse object detection .eim, and outputs the results over the camera feed as bounding boxes or circles for FOMO models

Write a Python script that watches the ./inbox folder for new CSV files, uploads each one to my Edge Impulse project as a training sample using the filename as the label, then moves it to ./processed

Write a shell script that uploads all .wav files in ./recordings to my Edge Impulse project, then starts a training job and waits for it to finish

Write a script that checks my Edge Impulse project every hour, and if a new training run has completed, exports the C++ library to ./build and sends a notification to a Slack webhook

Write a GitHub Actions workflow that triggers on push to main, uploads any new CSV files in ./data to Edge Impulse, retrains the model, and exports the C++ library as a build artifact

Write a GitHub Actions workflow that runs model testing after training and fails the build if accuracy drops below 90%

List all projects in my Edge Impulse account with their IDs and sample counts

Copy all training samples with label "idle" from project 1234 to project 5678

Compare the test accuracy of the current model in project 1234 versus project 5678 and show which is better

Export the best-performing model from project 1234 and import it into project 5678 as a version snapshot

Create a version snapshot of my Edge Impulse project with the description "before retraining with new data"

List all version snapshots for my project and show their descriptions and creation dates

Show me the model accuracy for each saved version in my project

Show a summary of my Edge Impulse project: sample counts per label, last training date, current accuracy, and deployment status

Generate a markdown report of my project's training history — one row per job with date, accuracy, and loss

Check whether my project has any samples that haven't been used in training (i.e., uploaded after the last training run)

Tell me how many samples were added to my project in the last 7 days, broken down by label

Create a report of all of the metadata in my project