AlphaStream India

AI-Powered Investment Intelligence for the Indian Investor

14 crore+ demat accounts. Most retail investors flying blind. AlphaStream India turns ET Markets data into actionable, money-making decisions.

Built for ET AI Hackathon 2026 - Problem Statement 6: AI for the Indian Investor.

---

What It Does

AlphaStream India is a real-time investment intelligence terminal that combines multi-agent signal detection, technical analysis overlays, NLQ analytics, and global market intelligence to surface opportunities other investors miss.

Core Features

Feature	Description
Tabbed Bloomberg Terminal	5-tab layout: Overview · Signals · Global Intel · Company · Portfolio - each focused on a different investor workflow
Opportunity Radar	AI monitors NSE/BSE filings, insider trades, FII/DII flows, chart patterns - surfaces signals with Alpha Score (0-100)
Technical Indicator Overlays	RSI(14) sub-chart + SMA20/SMA50 line overlays on TradingView candlestick, togglable per ticker
Fundamentals Panel	PE, PB, ROE, Dividend Yield, Market Cap, 52-week H/L from Groww API - wired live
Stock Screener	Filter Nifty 50 universe by sector, signal direction, and alpha score using DuckDB v_stock_screener view
Portfolio Manager	Add/remove NSE holdings, track real-time P&L, ₹ totals, horizontal bar chart by ticker
Watchlist	Persistent (localStorage) watchlist of up to 20 NSE tickers with live sentiment scores
Corporate Filings	BSE announcements (dividends, results, board meetings) per ticker - 7/30/90-day view
Anomaly Detection	Online ML (River HalfSpaceTrees) flags price/volume anomalies fed from 3mo NSE OHLCV
Global Market Intelligence	Crypto (BTC/ETH/SOL/XRP) + Currencies (INR/USD, DXY) + US Sector ETFs - India-impact annotated
Threat-aware News	Articles sorted by threat level (critical/warning/info) with sentiment distribution donut chart
Market ChatGPT Next Gen	NLQ agent with Text2SQL pipeline - grounded answers from real data, portfolio-aware, source-cited

---

Architecture

┌─────────────────────────────────────────────────────────────────────┐
│  Indian Data Sources                                                 │
│  NSE API · BSE API · FII/DII (NSDL) · Groww API · ET Markets RSS   │
│  FRED (macro) · WorldMonitor (global indices/commodities/crypto/FX)  │
└──────────────┬──────────────────────────────────────────────────────┘
               │
┌──────────────▼──────────────────────────────────────────────────────┐
│  Pathway Streaming Engine (Real-time <2s)                            │
│  News ingestion → Chunking → Embedding → Adaptive RAG               │
│  Threat classification (critical/warning/info) → DuckDB persist      │
└──────────────┬──────────────────────────────────────────────────────┘
               │
┌──────────────▼──────────────────────────────────────────────────────┐
│  Multi-Agent System (13 agents)                                      │
│  Sentiment · Technical · Risk · Decision · Pattern · Backtest        │
│  Filing · Flow · Insider · Chart · Report · Search · Anomaly         │
└──────────────┬──────────────────────────────────────────────────────┘
               │
┌──────────────▼──────────────────────────────────────────────────────┐
│  Analytics Layer (DuckDB)                                            │
│  fact_articles · fact_signals · fact_insider_trades · fact_filings   │
│  fact_fii_dii_flows · dim_stocks (Nifty 50)                         │
│  Views: v_stock_screener · v_signal_summary · v_sector_heatmap       │
└──────────────┬──────────────────────────────────────────────────────┘
               │
┌──────────────▼──────────────────────────────────────────────────────┐
│  NLQ Agent (LangGraph 7-node)                                        │
│  Guardrail → Enrich (web search) → Route → Text2SQL → Narrate        │
│  MCP Servers: market_data · signals · portfolio · search             │
└──────────────┬──────────────────────────────────────────────────────┘
               │
┌──────────────▼──────────────────────────────────────────────────────┐
│  React Terminal (5 tabs)                                             │
│  Overview: Candlestick+RSI/SMA · Recommendation · Fundamentals       │
│  Signals:  Screener · Opportunity Radar · Flow Chart                 │
│  Global:   Crypto/FX/Sectors · Fear&Greed · Macro · Commodities      │
│  Company:  Filings · News+Threat Badges · Watchlist                  │
│  Portfolio: Holdings · P&L Chart · Sector Allocation                 │
└─────────────────────────────────────────────────────────────────────┘

---

Tech Stack

Layer	Technology
Streaming	Pathway (real-time RAG, <2s latency)
LLM	Gemini 2.0 Flash (Vertex AI) + OpenRouter
Agents	LangChain + LangGraph (multi-agent orchestration)
NLQ	Text2SQL pipeline with guardrails + correction loop
MCP	FastMCP servers (market data, signals, portfolio)
Database	DuckDB (analytics), ChromaDB (vector search)
Backend	FastAPI + WebSocket + SSE streaming
Frontend	React 19, Zustand, Tailwind, Recharts, Framer Motion
Market Data	NSE API, BSE API, Groww API (TOTP), yfinance, ET Markets RSS

---

Quick Start

Prerequisites

• Python 3.11+ — backend runtime
• Node.js 18+ — frontend build tooling
• Google Cloud account with Vertex AI enabled (for Gemini LLM)
• API keys needed: NEWS_API_KEY from newsapi.org (required), GROWW_API_TOKEN (optional)
• Hardware: 4 GB+ RAM recommended

Step-by-Step Setup

Step 1: Clone & configure

git clone https://github.com/ridash2005/AlphaStream_Final.git
cd AlphaStream_India

Step 2: Backend setup

cd backend
uv sync
cp .env.example .env
# Edit .env — set GOOGLE_APPLICATION_CREDENTIALS, GCP_PROJECT_ID, and NEWS_API_KEY at minimum

Step 3: Initialize database & start backend

# Initialize DuckDB with Nifty 50 data
.venv/bin/python -m src.data.market_schema

# Start backend (cold start ~75s, then ready)
.venv/bin/python -m uvicorn src.api.app:app --host 0.0.0.0 --port 8000

Step 4: Frontend setup

cd ../frontend
npm install

Step 5: Start frontend

npm run dev

Step 6: Open the dashboard

Navigate to http://localhost:5173

Minimal Mode (No API Keys)

If you don't have API keys yet, AlphaStream still runs using cached/mock data. Set in backend/.env:

ENABLE_PATHWAY=false
NEWS_API_KEY=
GCP_PROJECT_ID=

The UI loads with cached article data and pre-computed recommendations — sufficient to evaluate the dashboard, agent outputs, and PDF report generation.

Environment Variables

See backend/.env.example for the full reference. Key variables:

# Required - GCP / Vertex AI (all LLM calls go through Gemini)
GOOGLE_APPLICATION_CREDENTIALS=./service-account.json
GCP_PROJECT_ID=your-project

# News APIs
NEWS_API_KEY=...

# Optional - Indian market data (enables fundamentals panel)
GROWW_API_TOKEN=...
GROWW_TOTP_SECRET=...

# Optional - enable Pathway streaming (adds 45s to startup)
ENABLE_PATHWAY=false

Troubleshooting

ImportError: No module named 'pathway' — Set ENABLE_PATHWAY=false in backend/.env. The system runs without Pathway using a synchronous fallback.

Port 8000 already in use — Run uvicorn src.api.app:app --port 8001 then set VITE_API_URL=http://localhost:8001 in frontend/.env.

Frontend loads blank — Check VITE_API_URL in frontend/.env points to the running backend (default: http://localhost:8000).

DuckDB error on startup — Run .venv/bin/python -m src.data.market_schema from the backend/ directory to initialise the database.

---

API Endpoints

Market Intelligence

Endpoint	Method	Description
/api/ohlcv/{ticker}	GET	OHLCV + optional RSI/SMA20/SMA50 (?indicators=true)
/api/fundamentals/{ticker}	GET	PE, PB, ROE, Div Yield, 52w H/L (Groww API)
/api/radar	GET	Top signals by Alpha Score
/api/screener	GET	Filter stocks by sector/direction/alpha (DuckDB view)
/api/patterns/{ticker}	GET	Chart pattern detection
/api/backtest/{ticker}/{pattern}	GET	Historical backtest (5yr)
/api/flows	GET	FII/DII flow analysis
/api/anomalies/{ticker}	GET	Price/volume anomaly detection (River ML)
/api/filings/{ticker}	GET	BSE corporate announcements
/api/portfolio	POST	Set user portfolio holdings
/api/portfolio/summary	GET	Live P&L summary

Global Market (WorldMonitor)

Endpoint	Method	Description
/api/global/indices	GET	NIFTY, SENSEX, S&P 500, DOW, etc.
/api/global/commodities	GET	Gold, Crude, Silver, Copper, etc.
/api/global/crypto	GET	BTC, ETH, SOL, XRP
/api/global/currencies	GET	INR/USD, DXY
/api/global/sectors	GET	12 US sector ETF returns
/api/global/fear-greed	GET	CNN Fear & Greed index
/api/global/macro	GET	FRED macro signals + verdict
/api/global/vix	GET	VIX volatility index
/api/global/geo-risk	GET	India geopolitical risk score

NLQ & Core

Endpoint	Method	Description
/api/nlq	POST	Natural language query (blocking)
/api/nlq/stream	GET/POST	NLQ with SSE streaming
/api/insights	GET	Ambient AI alerts
/recommend	POST	Full multi-agent recommendation
/ws/stream/{ticker}	WS	Real-time updates (rec, market, global)

---

Signal Types

Signal	Detection Method	Backtest Available
RSI Divergence	Price vs RSI direction mismatch	Yes (5yr)
MACD Crossover	Signal line cross + histogram	Yes (5yr)
Bollinger Breakout	Squeeze → expansion	Yes (5yr)
Volume Breakout	>2x 20-day average	Yes (5yr)
Golden/Death Cross	50 SMA vs 200 SMA	Yes (5yr)
Insider Buying	NSE SAST/PIT cluster detection	N/A
FII Streak	5+ day consecutive net buying	N/A
Material Filing	LLM classification of BSE/NSE filings	N/A

---

Author

Rickarya Das — @ridash2005

Built for ET AI Hackathon 2026 - Problem Statement 6

---

License

MIT