---
dataset_info:
  features:
  - name: messages
    list:
    - name: content
      dtype: string
    - name: role
      dtype: string
  splits:
  - name: train
    num_bytes: 3021564
    num_examples: 3043
  download_size: 1331200
  dataset_size: 3021564
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# TradeGuru Electrical Fault-Finding SFT Dataset

Synthetic training data for fine-tuning **Google Gemma 3 27B** as a specialised electrical fault-finding assistant for the **TradeGuru** iOS app.

## Dataset Details

- **Examples**: 3,043 conversation pairs
- **Format**: Gemma SFT (`messages` column with `user`/`model` roles)
- **Model responses**: Structured JSON matching the TradeGuru iOS app `ContentBlock` schema

## Source Knowledge

- 127-page professional electrical fault-finding guide covering domestic, commercial, and industrial domains
- 522 practical training summaries across 18 categories (testing, wiring, tools, earthing, safety, and more)

## Categories Covered (26)

Electrical basics, safety/PPE, LOTO, fault-finding process, testing equipment, CAT ratings, domestic (RCD faults, no power, breaker tripping, hot water), commercial (HVAC, emergency lighting, power distribution), industrial (motors, PLCs, sensors), wiring (switches, power points), earthing, cable work, tools, apprentice skills, 3-phase, roof work, switchboards.

## Response Format

Model responses are valid JSON objects with a `blocks` array containing typed content blocks:

```json
{"blocks": [
  {"type": "text", "content": "..."},
  {"type": "step_list", "title": "...", "steps": ["...", "..."]},
  {"type": "warning", "content": "..."},
  {"type": "callout", "content": "...", "style": "tip"}
]}
```

**Block types**: text, heading, step_list, warning, code, parts_list, regulation, callout, table, link

## Generation

- Generated via GoodWork swarm (Z.AI / NullClaw) with parallel agent orchestration
- Post-processed with validation pipeline: JSON schema checks, deduplication, compliance filtering
- Quality-filtered from 15,032 raw pairs down to 3,043 unique validated examples

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("THEARTISTISME/tradeguru-electrical-sft")
```