TradingNews.press — AI Financial News Analysis Platform

Overview

TradingNews.press is an AI-powered platform for automated financial news analysis and investment signal generation, developed from inception during my internship at Bauhinia AI Limited (affiliated with Prof. Kani Chen’s Lab at HKUST).

Technical Highlights

CRDR Workflow Orchestration

Architected a Channel→Relevant→Decision→ReAct workflow using LangGraph state machines with parallel processing:

• Channel: Extracts assets via JSON Schema-constrained outputs
• Relevant: Enriches context with market data
• Decision: Aggregates analyst signals from multiple agents
• ReAct: Refines low-confidence results through iterative reasoning

Multi-Agent System

Developed 7 specialized financial analysis agents with:

• Custom prompt engineering for financial domain expertise
• Pydantic output schemas for structured reasoning chains
• Production-grade error handling with fallback strategies and automatic retry logic

Text-Grad Optimization

Implemented a text-grad architecture featuring:

• Stock price tracking and prompt propagation computation graph
• LLM-as-a-judge framework for evaluating agent performance
• Learnable prompt components optimization across agents
• Data flywheel for continuous improvement

Impact

The platform is currently live and serving active users, providing real-time financial news analysis and investment signal generation.