Use Cases and Workflows

Real-world ML applications and end-to-end workflow examples, optimized for local VS Code development and automated execution.

NEO demonstrates practical ML solutions with reproducible workflows and clear artifact generation.

Pipeline Overview

NEO executes a full ML pipeline locally or in user-configured cloud environments:

Data Ingestion → Validation → Feature Engineering → Model Training → Evaluation → Deployment

By default, pipelines run locally inside your VS Code environment. Cloud resources are used only when explicitly configured.

Each stage includes automatic quality checks, error handling, and progress tracking.

ML Use Cases

Use Case	Type	Data Formats	Duration	Description
ETA prediction	Tabular and time series	CSV, Parquet	15-30 min	Predict arrival times from trip logs and temporal/geospatial features
Fine-tuning Whisper	Speech recognition	Audio and transcripts	2-6 hours	Domain-specific transcription via fine-tuning pre-trained models
Sleep stage prediction	Biomedical	Time series	30-60 min	Classify sleep stages from EEG or wearable sensor data
Network anomaly detection	Cybersecurity	Multi-source	15-30 min	Detect anomalous network traffic in real-time logs
Customer churn analysis	Business analytics	Tabular	15-30 min	Predict customer churn from transactional and demographic data

Durations vary based on dataset size, model choice, and local hardware.

ETA Prediction (Tabular and time series)

ETA prediction

Tabular and time series

Problem statement

Predict arrival time using trip logs, geographic coordinates, and temporal features.

Execution approach

1Ingest trip logs from CSV/Parquet

2Extract temporal features (hour, weekday, holidays)

3Engineer geospatial features (distance, bearing)

4Train gradient-boosted trees and neural networks

Generated artifacts

Cleaned dataset snapshot

Feature pipeline

Evaluation reports

Serialized model (ONNX or Pickle)

Fine-Tuning Whisper (Speech recognition)

Fine-tuning Whisper

Speech recognition

Problem statement

Fine-tune a pre-trained speech model (Whisper) for domain-specific transcription.

Execution approach

1Ingest audio files with transcripts

2Standardize sampling rates

3Apply data augmentation (noise, pitch)

4Transfer learning from Whisper checkpoint

Generated artifacts

Preprocessed audio dataset

Checkpoint artifacts (HF-compatible)

WER evaluation reports

Model card

Engineering Workflows

Debugging and error resolution

NEO analyzes stack traces and code logic to provide targeted fixes in VS Code.

”Check auth_middleware.py - why is the session token null after third redirect?”

Local data analysis

Perform automated exploratory data analysis on local files while keeping data private.

”Analyze sales_data.csv and report null distributions”

Open source contribution

Simplify contributions to unfamiliar codebases via code navigation, issue mapping, and PR suggestions.

”Which files need modification to fix GitHub issue #402?”

Best Practices

Provide clear context

Include dataset description, target metrics, and business goals

Iterate on results

Start broad, refine predictions and features based on insights

Validate production readiness

Request reproducible artifacts and monitoring configs before deployment

Monitor and maintain

Track drift, set alert thresholds, and schedule re-training

Next Steps

Quick Start

Get up and running in under 5 minutes

Features

Explore all capabilities

FAQ

Find answers to common questions