End-to-End ML Automation from Single Prompts

NEO acts as an autonomous ML engineer in VS Code, handling the full pipeline—data engineering, model training, deployment, and UI creation—without manual intervention. Prompt it with a task like "build a chat moderation pipeline to detect profanity and harmful text in messages," and it scans your workspace, creates a detailed task plan (e.g., generate synthetic data since none provided), and executes step-by-step. This replaces the need for separate data scientists, backend engineers, and DevOps roles, which typically make building agents a "nightmare" of data cleaning, feature engineering, hyperparameter tuning, and deployment.

Key to its reliability: before execution, NEO outlines stages like dataset engineering (schema definition, annotation guidelines for consistent labels), model selection (analyzes data to pick baseline classifier), training (splits train/validation sets, runs locally), evaluation (generates reports, logs metrics), API building (endpoints, serialization, requirements.txt), and frontend (interactive web UI for testing inputs like "Hey everyone how's the game going?" classified as clean vs. toxic text flagged with categories and confidence scores). All outputs land directly in your VS Code workspace as inspectable files (CSV with thousands of balanced rows covering profanity, hate speech, bullying, threats; training scripts; model weights), eliminating import/export hassles.

Use auto mode for self-checks and refinement passes if results fall short, or switch to pro mode for deeper logs and context retention in production workflows. Pause, review, interrupt, or stop anytime to retain control.

Local-First Execution with Cloud Integrations

NEO runs entirely locally on your machine for privacy—code, data, and encrypted credentials stay isolated per workspace, preventing context leakage across projects. Install free from VS Code marketplace, sign in with Neo account, open project folder, and go. No uploading repos to external environments.

Connect integrations like AWS S3 (pull real datasets), Hugging Face (models), Weights & Biases (experiment tracking with run logs), GitHub, or Kaggle via settings. If dependencies fail (e.g., CUDA issues, package versions), NEO inspects logs, adjusts setup, and recovers automatically—fixing common ML workflow breakers like environment mismatches.

Detailed real-time logs include timestamps, errors, recovery actions, and performance data, making processes transparent vs. black-box tools. For prototyping, light mode suffices; for serious work, pro mode adds control.

Broad Applicability and Real-World Value

Supports diverse workflows: tabular ML, forecasting, computer vision, OCR, speech, LLM fine-tuning, RAG systems, churn prediction, image models, retrieval pipelines, evaluation. Excels at "boring plumbing"—data prep, baseline training, debugging, shipping usable models—while top researchers handle novel architectures.

In the chat moderation demo without provided data, NEO generated synthetic CSV (multilingual, validated), trained/evaluated baseline, deployed real-time API, and built testable UI in one flow. Test inputs show accurate flagging (harmless: clean; toxic: harmful categories with scores). This delivers production-ready prototypes faster than manual efforts, especially for applied ML where 80% of time is non-research drudgery.

Trade-off: Best for practical engineering, not inventing new SOTA; requires VS Code and local Python env. Free credits via signup link make trialing low-risk.