GlobalLaunch-AI

Top Countries List Generated
ChatBot
Theoretical Graphs for Visualizations
Report page (with each country separate report)
Landing Page

🌟 Inspiration

Expanding a startup globally is a high-stakes decision, especially for early-stage teams lacking local insights. We were inspired by founders struggling to identify viable markets abroad. While tools exist to analyze domestic feasibility, none offer instant, data-driven international expansion guidance. We imagined a product that could blend semantic startup understanding with country-level indicators and deliver strategic advice instantly. That became GlobalLaunch-AI.

🚀 What it does

GlobalLaunch-AI is an AI-powered strategic expansion advisor. Here's what it does:

🧠 Detects sectors from business ideas or pitch decks using Gemini.
🌍 Shortlists the top 5 countries for expansion using semantic similarity and key macro indicators.
📊 Generates detailed AI-driven expansion reports for each country.
💬 Enables interactive querying through a chatbot that references real-time generated reports.

All of this is wrapped in a clean, responsive frontend with report browsing, dark mode, and dynamic chatbot support.

🛠️ How we built it

Backend Stack

Python + Flask for API orchestration
MongoDB for storing reports, semantic vectors, and metadata
Google Gemini (Generative + Embedding models) for:
- Sector detection
- Report generation
- Semantic similarity
- Chatbot responses
VertexAI for vector embeddings & semantic search

Frontend Stack

HTML/CSS + Tailwind + Vanilla JS
Full support for uploading PDFs, showing analysis cards, and toggling chat with AI
Dynamic pages (index.html, report.html) for streamlined UX

Infrastructure

PDF text extraction via PyMuPDF
Sector detection fallback logic with keyword classification
Embedding and indexing of country-sector profiles
Prompt-engineered multi-part report generation using structured JSON formatting

🧱 Challenges we ran into

⚙️ Prompt consistency with Gemini was a recurring issue, especially with JSON responses that needed parsing and validation.
🧠 Balancing heuristic vs LLM-based sector detection for reliability.
🧩 Merging chunked country data for semantic summaries required multiple retries and fallback logic.
💾 MongoDB index tuning for $vectorSearch was complex under time constraints.
🧪 Getting the chatbot to be relevant, concise, and grounded in the correct context was a delicate balance of prompt tuning.

🏆 Accomplishments that we're proud of

Built a fully functioning pipeline from raw idea to AI-generated, strategic country reports — end to end.
Designed a chatbot that intelligently answers regulatory or market-related questions using live, generated context.
Developed a modern, polished frontend with real-time feedback, dark mode, and semantic search filtering.
Efficiently embedded 100+ country-sector combinations and performed cross-sector vector ranking.

📚 What we learned

Prompt design for LLMs is an iterative and sometimes brittle process — retries and safety nets matter.
Embedding similarity can be a powerful tool when mixed with weighted indicators — hybrid scoring models win.
UX around "AI responses" must carefully manage user expectations and loading states.
Session management and auto-cleanup (via timed reset or page unloads) is essential in stateless AI applications.

🔮 What's next for GlobalLaunch-AI

🌐 Expand the database to include more granular regional/state-level indicators.
🧩 Add fine-tuned user controls (e.g., sector preference weights, regulatory strictness sliders).
📈 Build dashboards that visually compare countries on key metrics.
📤 Enable PDF report export and embedding insights in investor decks.
🤖 Replace hardcoded chatbot prompt with context-aware few-shot tuning.
📡 Possibly integrate real-world market APIs (e.g., World Bank, IMF, startup ecosystem APIs) for more real-time precision.