Skip to main content
Pioneer lets you fine-tune open-source LLMs and NER models on your own data, evaluate their performance, and serve predictions in production — without managing infrastructure. Connect via the Pioneer REST API, or use drop-in OpenAI and Anthropic-compatible endpoints.

Quick Start

Make your first API call in minutes. Get your API key, run inference, and fine-tune your first model.

API Reference

Full reference for every Pioneer endpoint — inference, training, datasets, evaluations, and projects.

Models

Browse available encoder (GLiNER) and decoder (LLM) models for fine-tuning and serverless inference.

Guides

Step-by-step walkthroughs for NER fine-tuning, LLM training, synthetic data generation, and more.

​
How Pioneer works

1

Choose a base model

Select from encoder models (GLiNER for NER/extraction) or decoder models (Qwen, Llama, DeepSeek, and more) depending on your task.
2

Upload or generate training data

Upload your labeled dataset or use Pioneer’s synthetic data generation to create training examples from scratch.
3

Start a training job

Submit a fine-tuning job via the API. Pioneer handles LoRA or full fine-tuning and reports F1, precision, and recall on completion.
4

Run inference

Call POST /inference with your training job ID to serve predictions. Your fine-tuned model is deployed on-demand — no cold-start setup required.
Pioneer also supports Adaptive Inference — a continuous improvement loop that automatically evaluates, retrains, and promotes model checkpoints based on live production traffic. See Adaptive Inference to learn more.

​
Key capabilities

NER Fine-tuning

Train GLiNER models on your entity types for extraction, classification, and structured JSON output.

LLM Fine-tuning

Fine-tune Qwen, Llama, DeepSeek, and other open-source LLMs using LoRA on your domain data.

Synthetic Data

Generate labeled training data for NER and classification tasks — no manual annotation required.

Agent Skills

Give your AI coding agent (Cursor, Claude Code) full Pioneer API knowledge with a single SKILL.md file.