Technical Architecture & System Flow
Designed & Engineered by Raj Puthawala
End-to-End System Topology
Framer Motion & Recharts
Upstash Redis Rate Limit
Groq / Llama 3.3 70B
FinBERT Sentiment NLP
EDGEIQ MICROSERVICE ROUTING MATRIX:
├── Finnhub API ───────> Quotes, News, Fundamentals, Insiders, Earnings, Options, Calendar
├── HuggingFace FinBERT ──> News Sentiment Classification (Bullish / Bearish / Neutral / Unavailable)
├── Groq / Llama 3.3 70B ─> Qualitative Research, Market Bias, Autopsy, Pre-Trade, Observations
└── Next.js Route Layer ─> Upstash Sliding Window Limit + Zod Input & Output Validation
1. Financial Data Pipeline
Connects to Finnhub APIs via serverless proxy handlers to fetch real quotes, company profiles, balance sheet metrics, insider Form 4 transactions, earnings surprise histories, and option chain snapshot distributions.
2. FinBERT Financial NLP Engine
Processes raw financial news headlines through HuggingFace's ProsusAI/finbert sentiment model. Returns probability distributions across Bullish, Bearish, and Neutral classifications with explicit fallback protection.
3. Groq Llama 3.3 70B AI Engine
Powers AI Stock Research Briefs, Trade Autopsies, Pre-Trade Risk Scorecards, Market Bias Summaries, and Market Observations. Operates with strict neutral research assistant persona prompts and untrusted data boundary delimiters.
4. Rate Limiting & Security Layer
Implements Upstash Redis sliding-window rate limiting (10 requests / 1 min per IP) with local in-memory fallback. Ensures API keys are strictly server-side and never exposed to the client bundle.