Automated investment teaser generation from company data — editable PowerPoint decks with native charts, anonymization, and full source citations.
deckAIn reads a markdown OnePager with company financials and produces a 5-slide investment teaser in under 3 minutes. The output is a fully editable .pptx with real Excel-backed charts, professional Kelp branding, and a companion citations document that traces every data point back to its source.
| Capability | Description |
|---|---|
| Native Charts | Excel-backed bar/line charts via python-pptx — click to edit in PowerPoint |
| Adaptive Extraction | Works with whatever financial data exists; missing EBITDA or PAT won't break the pipeline |
| Multi-Sector Intelligence | Pharma, Manufacturing, Technology, Services, Logistics, Healthcare, Consumer |
| Auto-Anonymization | Company names replaced with codenames; verified across all slides |
| Public Data Fusion | DuckDuckGo search for market context, trends, and competitive landscape |
| Image Sourcing | Unsplash API with compliance checks (logo/text detection via OCR) |
| Citations | Every claim maps to a markdown line or web URL in a .docx audit trail |
| Compliance Validation | Automated checks for chart editability, anonymization, branding, and slide count |
The pipeline generates presentations like this:
| Cover | Business Overview | Financials | Investment Highlights | Disclaimer |
|---|---|---|---|---|
| Slide 1 | Slide 2 | Slide 3 | Slide 4 | Slide 5 |
Output files per run:
output/CompanyName-OnePager/
├── Project_Codename_Investment_Teaser.pptx # 5-slide presentation
├── Project_Codename_citations.docx # Source audit trail
├── images/ # Sourced photos
└── run_YYYYMMDD_HHMMSS.log # Execution log
- Python 3.10+
- Google Gemini API key (required — free tier is sufficient)
- Unsplash API key (optional — improves image quality; DuckDuckGo is the fallback)
- Groq API key (optional — enables experimental visual review)
cd deckain
# Create and activate virtual environment
python -m venv .venv
.\.venv\Scripts\Activate.ps1 # Windows
# source .venv/bin/activate # macOS / Linux
# Install dependencies
pip install -r requirements.txt
# Configure API keys
copy .env.example .env # Windows
# cp .env.example .env # macOS / Linux
# Then edit .env and add your GOOGLE_API_KEYpython main.py <markdown_path> --sector <sector> --company <name>| Argument | Required | Description |
|---|---|---|
markdown_path |
Yes | Path to the company OnePager .md file |
--sector |
Yes | Industry — one of: Pharma, Manufacturing, Technology, Services, Logistics, Healthcare, Consumer |
--company |
No | Original company name (used to verify anonymization) |
# Pharma company
python main.py "..\Company Data\pharma-ind-swift\Ind Swift-OnePager.md" --sector Pharma --company "Ind Swift"
# Logistics company
python main.py "..\Company Data\logistics-gati\Gati-OnePager.md" --sector Logistics --company "gati"
# Technology company
python main.py "..\Company Data\technology-ksolves\Ksolves-OnePager.md" --sector Technology --company "ksolves"
# Entertainment / Services
python main.py "..\Company Data\entertainment-connplex\Connplex Cinemas-OnePager.md" --sector Services --company "connplex" ┌─────────────────────────────────────┐
│ Company OnePager.md │
└──────────────┬──────────────────────┘
│
┌────────────────────┼────────────────────┐
▼ ▼ ▼
┌─────────────────┐ ┌──────────────────┐ ┌─────────────────┐
│ 1. Financial │ │ 2. Chart Spec │ │ 5. Image │
│ Extraction │ │ Generation │ │ Sourcing │
│ (Gemini API) │ │ (Sector-aware) │ │ (Unsplash/DDG) │
└───────┬─────────┘ └───────┬──────────┘ └───────┬─────────┘
│ │ │
▼ │ │
┌─────────────────┐ │ │
│ 3. Content │ │ │
│ Generation │◄─────────┘ │
│ (Gemini API) │ │
└───────┬─────────┘ │
│ │
▼ │
┌─────────────────┐ │
│ 4. Citations │ │
│ Builder │ │
└───────┬─────────┘ │
│ │
└───────────────┬──────────────────────────┘
▼
┌──────────────────┐
│ 6. PPT Assembly │
│ (python-pptx) │
└───────┬──────────┘
▼
┌──────────────────┐
│ 7. Compliance │
│ Validation │
└───────┬──────────┘
▼
┌──────────────────┐
│ .pptx + .docx │
└──────────────────┘
7 stages execute sequentially:
| Stage | Module | What it does |
|---|---|---|
| 1 | extractors/private_data.py |
Uploads markdown to Gemini File API; extracts financials into a flexible Pydantic schema |
| 2 | generators/chart_data.py |
Builds native chart specs (revenue trend, margin, sector-specific) |
| 3 | generators/content.py |
Generates slide text, merges private + public data, applies anonymization |
| 4 | generators/citations.py |
Creates .docx linking every claim to markdown lines or web URLs |
| 5 | assets/image_sourcing.py |
LLM-generated search queries → Unsplash/DuckDuckGo → OCR compliance check |
| 6 | assemblers/slide_builder.py |
Assembles branded slides with sector-aware layout variants |
| 7 | validators/compliance.py |
Chart editability, anonymization, branding, citation, and slide count checks |
deckain/
├── main.py # CLI entry point
├── batch_generate.py # Run all companies at once
├── requirements.txt # Python dependencies
├── .env.example # Template for API keys
│
├── config/ # Configuration
│ ├── settings.py # Loads .env, defines paths and colors
│ ├── prompts.py # LLM prompt templates
│ ├── sector_config.py # Per-sector KPIs, certs, terminology
│ └── sector_prompts.py # Sector-specific prompt augmentation
│
├── extractors/ # Stage 1: Data extraction
│ ├── private_data.py # Gemini-based financial extraction
│ ├── public_data.py # DuckDuckGo web scraping
│ ├── markdown_parser.py # Markdown section parser
│ └── schemas.py # Pydantic models for all stages
│
├── generators/ # Stages 2-4: Content creation
│ ├── chart_data.py # Native chart specifications
│ ├── content.py # Slide text generation
│ ├── citations.py # Source documentation builder
│ └── layout_engine.py # Experimental generative layouts
│
├── assets/ # Stage 5: Images and icons
│ ├── image_sourcing.py # Unsplash + DuckDuckGo image search
│ ├── image_compliance.py # OCR-based logo/text detection
│ ├── icon_helper.py # SVG → PNG icon management
│ ├── pillow_icons.py # Pillow-generated icons and cert badges
│ └── icons/ # Cached SVG/PNG icons
│
├── assemblers/ # Stage 6: PowerPoint assembly
│ ├── slide_builder.py # Builds all 5 slides
│ ├── ppt_template.py # Kelp-branded template (header/footer)
│ ├── native_charts.py # Excel-backed chart creation
│ └── components.py # Reusable visual components
│
├── validators/ # Stage 7: Quality assurance
│ ├── compliance.py # Automated compliance scoring
│ ├── groq_reviewer.py # LLM-based visual review (optional)
│ ├── slide_optimizer.py # Layout adjustments
│ └── visual_reviewer.py # Slide rendering for review
│
├── utils/ # Shared utilities
│ ├── llm_client.py # Gemini API wrapper with retries
│ ├── logger.py # Structured logging
│ ├── checkpoint.py # Pipeline checkpoint/resume
│ └── ppt_renderer.py # Slide → image rendering
│
└── output/ # Generated presentations
All settings live in .env (copy from .env.example):
# Required
GOOGLE_API_KEY=your_gemini_api_key
GEMINI_MODEL=gemini-3-flash-preview
# Optional — improves image quality
UNSPLASH_ACCESS_KEY=your_unsplash_key
# Optional — visual feedback loop
GROQ_API_KEY=your_groq_key
GROQ_MODEL=meta-llama/llama-4-maverick-17b-128e-instruct
# Feature flags
ENABLE_WEB_SEARCH=true
ENABLE_CITATIONS=true
ENABLE_COMPLIANCE_CHECKS=true
ENABLE_GROQ_REVIEW=false
USE_LLM_FOR_SLIDES=falseEach sector customizes KPIs, certifications, image queries, and slide terminology:
| Sector | Key Metrics | Certifications | Focus Areas |
|---|---|---|---|
| Pharma | ANDA count, R&D spend, export % | WHO-GMP, US FDA, CEP | API portfolio, therapeutic areas, regulatory filings |
| Manufacturing | Capacity utilization, facility count | IATF 16949, ISO 9001 | Production lines, quality, automotive/industrial |
| Technology | Developer count, ARR, client retention | ISO 27001, CMMI | SaaS metrics, tech stack, platform capabilities |
| Services | Client count, geographic reach | ISO 9001, CMMI | Service breadth, customer relationships |
| Logistics | Fleet size, ton-km, hub count | ISO 14001, AEO | Network density, utilization, tracking |
| Healthcare | Bed count, patient volume | NABH, JCI | Clinical specialties, outcomes |
| Consumer | SKU count, D2C %, repeat rate | FSSAI, BIS | Brand portfolio, distribution reach |
| Metric | Typical Value |
|---|---|
| End-to-end time | 60–120 seconds |
| Gemini API calls | 5–6 per teaser |
| Token usage | 30,000–50,000 total |
| Cost | Free tier sufficient (1,500 req/day) |
| Output | 5 slides + citations .docx |
| Choice | Rationale |
|---|---|
| Gemini + File API | Upload markdown once, reuse across prompts — saves tokens |
| python-pptx | Creates real PowerPoint objects; charts stay editable |
| Pydantic schemas | Type-safe validation with flexible optional fields |
| DuckDuckGo | Free web search, no API key required |
| Unsplash | License-clear professional photography |
| Sector configs | Single source of truth for industry-specific behavior |
MIT
Built for the Kelp M&A automation challenge (AI-ML GC IIT Bombay) — Jan - Feb 2026.