An Open-Source FinTerms Engine Powered by DeepSeek-V3
FinTerms is an auto-generated financial dictionary that combines prompts engineering with AI-native design, delivering:
✨ Structured Terminology
Markdown-native formatting with strict bilingual alignment (CN/EN).
🤖 DeepSeek-V3 Powered
Leveraging the cost-efficient architecture of DeepSeek's latest LLM for sustainable content generation at industry-leading rates:
- Input: $0.014-0.14 per 1M tokens
- Output: $0.28 per 1M tokens
- 64K context window
📚 Structured Knowledge Architecture
- Dual-language synchronized definitions and explanations
- Historical context and origin tracking
- Practical application scenarios and risk notes
- Regulatory and market significance analysis
- Structured key takeaways for quick reference
- LLM-reviewed entries based on Investopedia's Daily FinTerms¹
¹ The original word list is from Investopedia's Daily FinTerms. It was filtered and expanded by Gemini Advanced Experimental 2.0, then reviewed by DeepSeek-V3 using this code: meta/dict_review.py
Cost estimate based on a 500-word bilingual entry (~2000 tokens): input cost $0.00028 + output cost $0.00056 ≈ $0.00084 per entry Reference: DeepSeek-V3 API Pricing
Update/Create your .env file with your OpenAI API key and endpoint:
OPENAI_API_KEY=<your_openai_api_key>
OPENAI_ENDPOINT=<your_deepseek_endpoint>Run the script:
python3.10 -m venv .venv
source .venv/bin/activate
pip install python-dotenv openai
python gen.py --dict # generate all entries
python gen.py --index # generate all entries🚀 High Performance
- Multi-threaded term generation
- Smart retry mechanism with exponential backoff
- Efficient file handling with thread-safe operations
🎯 Smart Organization
- Automatic A-Z category sorting
- Special handling for numeric/symbol terms
- Sanitized filename generation for cross-platform compatibility
📊 Progress Tracking
- Real-time generation progress monitoring
- Detailed completion statistics
- Timestamp-based update tracking
🛠Developer Friendly
- CLI interface with --dict and --index options
- Template-based content generation
- Configurable via environment variables
- Automatic index generation with Markdown linking
🔄 Robust Processing
- Comprehensive error handling
- Duplicate entry detection
- Content validation for generated terms
- Auto-creation of category directories
- Currently DeepSeek-V3 is experiencing high traffic limitations, making multi-threading ineffective
- Looking for cost-effective hosting solutions to:
- Deploy self-hosted inference
- Support multiple API endpoints
- Enable full multi-threading capabilities with several API endpoints