Synthetic data

Bootstrap a training set with a model. Describe the task and get several diverse, correctly-formatted JSONL examples you'd review and expand. Synthetic data won't beat real data, but it gets a dataset off the ground.

README Try it Source

Showcase — Synthetic data

Run

bash bootstrap-secrets.sh              # reads ../../../../.env, writes secrets/
docker compose up --build              # default: PROVIDER=openai

Open http://localhost:3000. Gemini: PROVIDER=gemini docker compose up --build.

What's where

backend/ai_openai.py / backend/ai_gemini.py — generate JSONL examples for a task spec.
frontend/app/page.tsx — task description in, JSONL examples out.

Stop

docker compose down

Run locally

Download the project as a ZIP and run it with Docker. Brings up a FastAPI backend + Next.js frontend on localhost:3000.

Download synthetic-data.zip

unzip synthetic-data.zip
cd synthetic-data
bash bootstrap-secrets.sh   # one-time: pulls API keys into ./secrets
docker compose up --build   # default provider: openai
# or:  PROVIDER=gemini docker compose up --build

Type some input, pick a provider, and run the same code shown in Source against the live API. Sign-in required.

Input

Provider

Output

The same modules the Run button hits. The whole project (frontend, Dockerfile, compose) is in the ZIP under README.

`backend/ai_openai.py`

"""Showcase 3 (OpenAI): generate training data with a model.

The chicken-and-egg of fine-tuning: you need hundreds of examples and you have a
handful. A capable model can bootstrap them. Describe the task and get several
diverse, correctly-formatted JSONL training examples you'd review and then
expand. Synthetic data won't beat real data, but it gets a dataset off the ground.
"""
from openai import OpenAI

_client = OpenAI()

_MODEL = "gpt-5.4-nano"

_PROMPT = (
    "Generate 5 diverse fine-tuning examples for the task the user describes. "
    "Output JSONL, one object per line, each: "
    '{"messages": [{"role": "user", "content": "..."}, {"role": "assistant", "content": "..."}]}. '
    "Vary the inputs realistically. Output ONLY the JSONL, no prose."
)


def run(spec: str) -> str:
    s = spec.strip()
    if not s:
        return "Describe the task you want training data for (e.g. 'classify support tickets by urgency')."
    response = _client.responses.create(model=_MODEL, instructions=_PROMPT, input=[{"role": "user", "content": s}])
    return response.output_text

`backend/ai_gemini.py`

"""Showcase 3 (Gemini): generate training data with a model.

Same synthetic-example generation, on Gemini.
"""
import os

from google import genai
from google.genai import types

_client = genai.Client(api_key=os.environ["GEMINI_API_KEY"])

_MODEL = "gemini-3.1-flash-lite"

_PROMPT = (
    "Generate 5 diverse fine-tuning examples for the task the user describes. "
    "Output JSONL, one object per line, each: "
    '{"text_input": "...", "output": "..."}. '
    "Vary the inputs realistically. Output ONLY the JSONL, no prose."
)


def run(spec: str) -> str:
    s = spec.strip()
    if not s:
        return "Describe the task you want training data for (e.g. 'classify support tickets by urgency')."
    response = _client.models.generate_content(
        model=_MODEL,
        contents=[types.Content(role="user", parts=[types.Part(text=s)])],
        config=types.GenerateContentConfig(system_instruction=_PROMPT),
    )
    return response.text or ""