Tech Stack

​

Overview

┌─────────────────────────────────────────────────────────────┐
│                         FRONTEND                            │
│  React 19 + TypeScript + Tailwind CSS + Vite               │
└────────────────────┬────────────────────────────────────────┘
                     │ HTTP / WebSocket
┌────────────────────┴────────────────────────────────────────┐
│                         BACKEND                             │
│  FastAPI + Uvicorn + Pydantic                               │
└────────────────────┬────────────────────────────────────────┘
                     │
        ┌────────────┼────────────┐
        │            │            │
┌───────▼──────┐ ┌──▼─────┐ ┌───▼────────┐
│ PostgreSQL   │ │ Redis  │ │  AWS S3    │
│ + pgvector   │ │ Cache  │ │  Storage   │
└──────────────┘ └────────┘ └────────────┘
                     │
        ┌────────────┼────────────────────┐
        │            │                    │
┌───────▼──────┐ ┌──▼──────────┐ ┌──────▼────────┐
│  Cerebras    │ │   OpenAI    │ │    Tavily     │
│  Qwen-3-235B │ │ GPT-5-nano  │ │  News Search  │
└──────────────┘ └─────────────┘ └───────────────┘

​

Backend Stack

​

Web Framework

• Automatic OpenAPI/Swagger documentation
• Built-in request validation with Pydantic
• Native async/await support for streaming responses
• Type hints for IDE support and runtime validation

• High-performance async server
• WebSocket support for streaming
• Production-ready with standard configuration

• Lightweight and fast
• WebSocket and streaming support

​

Database & Storage

• Stores transcript chunks, 10-K chunks, and metadata
• JSONB support for flexible metadata storage
• Robust ACID guarantees for data consistency

• Efficient storage of 384-dimensional embeddings
• Native PostgreSQL vector operations
• Hybrid search combining vector similarity with traditional SQL

• Used for financial data analysis and screening
• Fast aggregations over company fundamentals

• Full transcript text and SEC filing documents
• Cost-effective storage for large documents
• Separation of embeddings (PostgreSQL) from full text (S3)

• Session caching and rate limiting
• Optional dependency for production deployments

​

AI/ML Stack

• Cerebras (primary) - Qwen-3-235B-A22B-Instruct-2507
  • Fast inference for question analysis, response generation, evaluation
  • Configured via CEREBRAS_API_KEY
• OpenAI (fallback) - GPT-5-nano-2025-08-07
  • Backup LLM provider
  • Configured via OPENAI_API_KEY
• Provider Selection - Configurable via llm_provider in rag/config.py (“cerebras” | “openai” | “auto”)

• sentence-transformers - all-MiniLM-L6-v2 model
  • 384-dimensional dense vectors
  • Optimized for semantic similarity search
  • Lightweight and fast inference

• LangChain - Orchestration and utilities
  • Text splitters for chunking transcripts
  • Community integrations for data sources
  • Core abstractions for LLM interactions

​

Search & Retrieval

• Vector Search (70% weight) - Semantic similarity via pgvector
• Keyword Search (30% weight) - TF-IDF for exact term matching
• Cross-encoder Reranking - Re-scores results for relevance (SEC agent)

• Earnings Transcripts - Hybrid vector + keyword search (search_engine.py)
• SEC 10-K Filings - Specialized retrieval agent with section routing (sec_filings_service_smart_parallel.py)
• Real-Time News - Tavily API integration (tavily_service.py)

​

Authentication & Security

• JWKS-based JWT verification (RS256)
• Secure token validation
• Configured via CLERK_SECRET_KEY and CLERK_PUBLISHABLE_KEY

• RS256 signature verification
• Required for Clerk integration

• python-jose, passlib, Authlib, Google OAuth libraries
• Retained for backward compatibility

​

Python Dependencies

# Core Framework
fastapi
uvicorn[standard]
starlette

# Database
asyncpg==0.30.0          # Async PostgreSQL driver
psycopg2-binary           # Sync PostgreSQL driver
SQLAlchemy                # ORM

# Authentication
PyJWT[crypto]             # Clerk JWT verification

# AI/ML
openai                    # OpenAI API client
cerebras-cloud-sdk        # Cerebras API client
langchain==0.3.18         # RAG orchestration
langchain-community==0.3.17
langchain-core==0.3.34
langchain-openai
langchain-text-splitters==0.3.6
sentence-transformers==3.4.1  # Embeddings
tiktoken==0.8.0           # Token counting

# Data Processing
pandas
numpy==2.2.5

# Utilities
pydantic==2.10.6          # Data validation
tavily==1.1.0             # News search
redis==5.2.1              # Caching
tenacity==9.0.0           # Retries
boto3>=1.35.0             # AWS S3

# Monitoring
logfire[fastapi]
logfire[asyncpg]

​

Frontend Stack

​

Core Framework

• Latest version with improved hooks and concurrent features
• Component-based architecture
• Virtual DOM for performance

• Compile-time type checking
• Enhanced IDE support
• Reduced runtime errors

• Fast hot module replacement (HMR)
• Optimized production builds
• Native ES modules support

​

UI & Styling

• Rapid UI development
• Consistent design system
• Minimal bundle size with tree-shaking

• Prose styles for markdown content
• Responsive text scaling

• Smooth transitions and animations
• Gesture support
• Layout animations

• Modern, consistent icon set
• Tree-shakeable for small bundle size

​

Routing & Markdown

• Nested routes and layouts
• Programmatic navigation
• URL parameter handling

• GitHub-flavored markdown support
• Code syntax highlighting
• Custom component rendering

• Tables, task lists, strikethrough
• Auto-linking

• Allows custom HTML elements in markdown

​

Authentication & Analytics

• Pre-built sign-in/sign-up components
• User management
• Session handling

• User behavior tracking
• Feature flags
• Session replay

​

Frontend Dependencies

{
  "dependencies": {
    "@clerk/clerk-react": "^5.60.0",
    "@posthog/react": "^1.8.0",
    "framer-motion": "^12.29.2",
    "lucide-react": "^0.563.0",
    "react": "^19.2.0",
    "react-dom": "^19.2.0",
    "react-markdown": "^10.1.0",
    "react-router-dom": "^7.13.0",
    "tailwindcss": "^4.1.18"
  },
  "devDependencies": {
    "@vitejs/plugin-react": "^5.1.1",
    "typescript": "~5.9.3",
    "eslint": "^9.39.1",
    "vite": "^7.2.4"
  }
}

​

External Services

​

Required APIs

​

Optional APIs

​

Development Tools

• Tenacity - Retry logic with exponential backoff
• Logfire - Observability for FastAPI and asyncpg
• Pydantic Settings - Environment variable management

• ESLint - Code linting
• TypeScript ESLint - TypeScript-specific linting rules
• Vite - Development server with HMR

​

Performance Optimizations

1. Async/Await - Non-blocking I/O throughout backend
2. Streaming Responses - Real-time answer generation and progress updates
3. Parallel Execution - Multi-ticker queries run concurrently
4. Connection Pooling - Efficient database connection management (asyncpg)
5. Vector Indexing - Fast similarity search with pgvector IVFFlat indexes
6. Hybrid Search - Combines semantic and keyword search for better recall
7. Cross-encoder Reranking - Improves precision for SEC filings
8. Redis Caching - Reduces redundant database queries (optional)

​

Next Steps