sinabis-group/sinabis-10k-synthetic-pptx

Sinabis Green Energy – Synthetic Document Dataset (10k PPTX)

10,100 business presentations (.pptx) from the everyday operations of a fictional group of Sinabis companies (Sinabis Green Energy, Sinabis Bank, Sinabis Logistics, Sinabis Electric …) – every organisation in the data carries the Sinabis name.

This is the 10k edition of sinabis-1m-synthetic-pptx: the first 10,100 presentations of the 1M dataset (idx 0 to 10,100). File names, folder paths and manifest records are identical to the 1M dataset, and this edition is the first part of the 100k edition, so all editions can be used side by side.

Every person, company, address, phone number, financial figure and project in this dataset is purely fictional. Deck contents come from LLM-written blueprints (Gemma 4 26B-A4B) in the target language: an outline per deck, then fully written slides in blocks of seven (6–45 slides per deck), each with layout, content and speaker notes. Slides are rendered with python-pptx in 17 layouts (title, agenda, section, bullets, two/three columns, KPI tiles, table, native chart, timeline, process, quote, comparison, team, image + text, FAQ, closing) and deck-level design variants (16:9/4:3, background, title position, accent, logo position, footer, card style, fonts).

All figures in this dataset card were computed directly from the files in this edition (as of 23 Oct 2026). They are not target estimates.

At a glance

Metric	Value
Total Presentations	10,100 .pptx presentations (1.03 GB uncompressed, 857 MB in 1 ZIP archive)
Total Slides	185,738 slides (average 18.4 slides per presentation, 6–45)
Aspect Ratio	16:9 widescreen for 9,088 decks (90.0 %), 4:3 for 1,012 decks (10.0 %)
Native Charts & Tables	7,984 decks (79.0 %) with native charts, 8,735 (86.5 %) with tables
Speaker Notes	Ø 1,398 words of speaker notes per presentation
Document Themes	20 distinct business themes in 5 families
Time Span	2011-05-19 to 2026-06-30 (15.1 years of corporate evolution)
Languages	24 languages (EN 81.6 %, JA 4.9 %, rest across 22 languages)
Issuing Organisations	1,957 fictional companies across 33 countries
Partner Organisations	903 distinct named recipients (partners, suppliers, clients); 78.3 % of decks are internal

What it is about

Sinabis Green Energy (headquartered in Austin, Texas, fictional) develops, constructs, operates and finances utility-scale renewable energy infrastructure:

• Solar & Storage: Rooftop PV, floating solar parks, utility-scale bifacial tracker arrays, and multi-megawatt battery energy storage systems (BESS).
• Wind & Microgrids: Onshore and offshore wind farms, smart grid controllers, SCADA power plant automation.
• Clean Transport & Charging: Mega-watt EV fleet charging hubs, autonomous campus shuttles, battery swap logistics.
• Circular Resources: Industrial water treatment, ocean plastic recovery, and sustainable building engineering.

Business units and themes represented

The 20 presentation themes cover the complete operational spectrum of modern industrial enterprises:

Theme	Count	Share	Description
corporate_presentation	651	6.4 %	Company overview, vision, business areas, global presence and core KPIs
project_kickoff	640	6.3 %	Project kick-off presentation with milestone plan, work packages, budgets and risk matrix
product_catalog	622	6.2 %	Product overview with storage series (C&I, utility), efficiency ratings and warranties
project_review	612	6.1 %	Construction and project progress report with actual costs, earned value and next steps
budget_review	588	5.8 %	Financial and budget review with target/actual cost-centre variances and forecast
sales_pitch	580	5.7 %	Customer presentation with ROI calculation, savings potential and PPA transformation concept
board_deck	515	5.1 %	Board report with strategic goals, segment results, investment plan and governance
maintenance_briefing	515	5.1 %	O&M service briefing with availability figures, maintenance packages and 5 safety rules
supply_chain	509	5.0 %	Global sourcing strategy with OTIF delivery performance, buffer stocks and LkSG supply chain audits
system_architecture	507	5.0 %	Technical system architecture for power plant control, BESS and grid integration
partnership_deck	499	4.9 %	Strategic partnership with R&D goals, joint go-to-market and aftersales SLA
hse_briefing	499	4.9 %	Occupational safety briefing with golden rules, mandatory PPE and stop-work authority
strategy_roadmap	492	4.9 %	Multi-year transformation roadmap with innovation clusters, capex and target states
case_study	492	4.9 %	Customer reference with load profile optimisation, peak shaving and testimonial feedback
esg_report	437	4.3 %	Sustainability report with scope 1/2/3 emissions, occupational safety and biodiversity
quality_audit	408	4.0 %	Quality audit report under ISO 9001 with CAPA corrective actions and top ratings
compliance_training	394	3.9 %	Code of conduct, anti-corruption, gifts policy and whistleblowing portal
townhall_meeting	393	3.9 %	Group-wide quarterly all-hands meeting with business-unit updates, financial highlights and Q&A
incident_report	379	3.8 %	Incident investigation (RCA) using the 5-why method with corrective actions and fleet lessons learned
employee_onboarding	368	3.6 %	Welcome deck for new employees with culture, tools, buddy and 90-day plan

Language distribution

The language of each presentation is determined by the issuing organisation's jurisdiction and recorded in the manifests:

Language	Presentations	Share
English (en)	8,245	81.6 %
Japanese (ja)	497	4.9 %
Chinese (zh)	446	4.4 %
German (de)	167	1.7 %
Portuguese (pt)	106	1.0 %
Spanish (es)	66	0.7 %
French (fr)	55	0.5 %
Italian (it)	49	0.5 %
Dutch (nl)	44	0.4 %
Swedish (sv)	41	0.4 %
Indonesian (id)	39	0.4 %
Korean (ko)	36	0.4 %
Hungarian (hu)	35	0.3 %
Polish (pl)	35	0.3 %
Finnish (fi)	33	0.3 %
Romanian (ro)	26	0.3 %
Danish (da)	26	0.3 %
Czech (cs)	25	0.2 %
Turkish (tr)	24	0.2 %
Norwegian (nb)	23	0.2 %
Greek (el)	23	0.2 %
Ukrainian (uk)	22	0.2 %
Vietnamese (vi)	21	0.2 %
Russian (ru)	16	0.2 %

Design DNA & Visual Hierarchy

Each presentation adheres to strict corporate design standards matching the Sinabis document family:

• 16:9 Widescreen and 4:3 Canvas: 90.0 % widescreen, 10.0 % classic 4:3; professional layout with precise millimeter typography and safe margins.
• Embedded Vector/Raster Logos: Crisp branding matching the company's industry and country.
• Executive Header & Footer Chrome: Category kicker, bold titles, subtitle hierarchy, date, confidentiality level, and slide counter (Slide X of Y).
• Diverse Slide Archetypes:
  • Title Slide: Brand-colored or dark-mode hero banner with corporate metadata.
  • KPI Dashboard: Multi-tile metrics with key values, trend badges and takeaway boxes.
  • Structured Column Cards: 3-column modular cards with priority badges and bullet streams.
  • Milestone Timelines: Horizontal multi-step roadmaps from project kickoff to commissioning.
  • Styled Tables: Crisp bordered tables with highlighted headers and sum rows.
  • Takeaway / Closing Slides: Executive call-to-action with verified contact details.

Quick start

The Parquet file under data/ holds one row per presentation: all manifest fields plus the extracted text, so the dataset can be browsed in the Dataset Viewer above and loaded directly:

from datasets import load_dataset

ds = load_dataset("sinabis-group/sinabis-10k-synthetic-pptx", split="train")
print(ds[0]["title"], ds[0]["slides"], "slides")
print(ds[0]["text"])     # slide text
print(ds[0]["notes"])    # speaker notes

The text column holds the slide text, slide by slide, each introduced by [Slide N]. The notes column holds the speaker notes in the same format; they are often longer than the slides themselves. The original .pptx files with layouts, charts and tables are in the ZIP archive.

Downloads

File	Size	Contents
data/train-*.parquet (1 file)	70 MB	Manifest fields + text (slides) + notes (speaker notes), one row per presentation (Dataset Viewer, load_dataset)
Sinabis_10k_pptx.zip	857 MB	10,100 PPTX files + manifest (1.03 GB uncompressed)

Sinabis_10k_pptx_manifest.zip (3.4 MB) holds only the JSON Lines manifest records for all 10,100 presentations, for a quick look at the metadata without downloading the decks (see "Metadata & Manifest" below).

Unpacking

Both archives share one directory structure. The presentation archive already contains the manifest, so unzip -o simply overwrites identical files:

hf download sinabis-group/sinabis-10k-synthetic-pptx --repo-type dataset --local-dir sinabis-10k-pptx
cd sinabis-10k-pptx
for z in Sinabis_10k_pptx*.zip; do unzip -qo "$z"; done

Folder structure

In the repository:

.
├── README.md
├── data/
│   └── train-*.parquet                 (rows for the Dataset Viewer and load_dataset)
├── Sinabis_10k_pptx.zip             → pptx/ + manifest/
└── Sinabis_10k_pptx_manifest.zip    → manifest/

After unpacking:

sinabis-10k-pptx/
├── pptx/
│   ├── board_deck/
│   │   ├── 0000/
│   │   │   ├── 0000000_PRS-2016-84085.pptx
│   │   │   └── ...
│   │   └── ... (0000 to 0010)
│   ├── project_kickoff/
│   ├── esg_report/
│   └── ... (20 themes)
└── manifest/
    ├── pb00000.jsonl
    ├── ...
    └── pb00020.jsonl

Metadata & Manifest

Each pbNNNNN.jsonl contains 500 records (the last shard 100), one per presentation. A real record from this dataset (slide_titles, the list of all slide titles, omitted for brevity):

{
  "idx": 0,
  "file": "pptx/board_deck/0000/0000000_PRS-2016-84085.pptx",
  "format": "pptx",
  "theme": "board_deck",
  "issuer_id": "painting-0-8",
  "issuer": "Sinabis Painters Bangkok Co., Ltd.",
  "issuer_country": "th",
  "recipient": "",
  "recipient_kind": "internal",
  "lang": "en",
  "date": "2016-01-23",
  "number": "PRS-2016/84085",
  "currency": "THB",
  "title": "Environmental Compliance & Sustainability Roadmap",
  "bytes": 93708,
  "family": "management",
  "blueprint_id": "board_deck_en_00031",
  "length_class": "M",
  "slides": 16,
  "layouts": {
    "title": 1,
    "agenda": 1,
    "section": 3,
    "kpi": 1,
    "two_column": 1,
    "comparison": 1,
    "three_column": 1,
    "chart": 1,
    "process": 1,
    "timeline": 1,
    "table": 1,
    "quote": 1,
    "faq": 1,
    "closing": 1
  },
  "charts": 1,
  "tables": 1,
  "notes_words": 1415,
  "confidentiality": "Confidential",
  "layout_format": "16:9",
  "layout_background": "white",
  "layout_title_position": "top_left",
  "layout_accent": "none",
  "layout_logo_position": "none",
  "layout_footer": "number_only",
  "layout_card_style": "top_border",
  "layout_font": "Tahoma",
  "layout_slide_layouts": "agenda,chart,closing,comparison,faq,kpi,process,quote,section,table,three_column,timeline,title,two_column"
}

Besides the identifying fields (idx, file, theme, issuer_id, issuer, issuer_country, recipient, recipient_kind, lang, date, number, currency, title), each record carries:

• family, blueprint_id, length_class: Theme family, source blueprint (<theme>_<lang>_<n>) and length class (S/M/L/XL)
• slides, layouts: Slide count and number of slides per slide layout
• charts, tables, notes_words: Native charts, tables and words of speaker notes
• confidentiality: Confidentiality label printed on the slides
• layout_*: Deck-level design decisions (format, background, title position, accent, logo position, footer, card style, font, slide layouts used)
• slide_titles: Titles of all slides in order
• bytes: Exact byte size of the .pptx file

How to use this dataset

Python (python-pptx)

from pptx import Presentation

prs = Presentation("0042100_BRD-2024-0812.pptx")
print(f"Total slides: {len(prs.slides)}")

for idx, slide in enumerate(prs.slides, 1):
    for shape in slide.shapes:
        if shape.has_text_frame:
            for p in shape.text_frame.paragraphs:
                if p.text.strip():
                    print(f"Slide {idx}: {p.text}")

Multimodal Vision & OCR Training

The slides feature high-contrast, structured text, tables, and KPI cards, making them ideal for training vision-language models (VLMs) on slide layout analysis, chart-to-text, and visual document reasoning.

License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license. You are free to share and adapt the material for any purpose, including commercial applications, provided appropriate credit is given:

Sinabis GmbH (2026). Sinabis Green Energy – Synthetic Document Dataset (10k PPTX). https://huggingface.co/datasets/sinabis-group/sinabis-10k-synthetic-pptx

Citation

@misc{sinabis_green_pptx_10k_dataset_2026,
  author = {{Sinabis GmbH}},
  title  = {Sinabis Green Energy -- Synthetic Document Dataset (10k PPTX)},
  year   = {2026},
  url    = {https://huggingface.co/datasets/sinabis-group/sinabis-10k-synthetic-pptx}
}