Back to Blog
Engineering

Under the Hood: How Stoxmine Processes 2.8 Million Data Points Daily

A technical deep dive into the engineering behind Stoxmine's real-time scoring pipeline — from NSE tick data to your Confidence Score.

SE

Stoxmine Engineering

Engineering Team

·8 July 2026·16 min read

Every trading session, Stoxmine ingests, processes, scores, and delivers intelligence on 5,000+ stocks in under 90 seconds. Here's how.

The Data Pipeline

Our pipeline runs in 4 stages, each optimised for throughput and latency:

Stage 1: Ingestion (0–15 seconds)

We pull data from multiple sources simultaneously:

  • Price & volume: Real-time tick data from NSE & BSE via licensed API feeds
  • Corporate data: Quarterly results, shareholding patterns, corporate actions from exchange filings
  • Institutional flows: Bulk/block deal data, FII/DII segment-wise data from NSDL & CDSL
  • News: RSS feeds and API integrations from 50+ Indian financial media outlets

Total daily ingestion: 2.8 million+ data points across all sources.

Stage 2: Normalisation & Feature Engineering (15–45 seconds)

Raw data is cleaned, normalised, and transformed into 127 quantitative features per stock. This includes:

  • Z-score normalisation for cross-sector comparison
  • Rolling window calculations (5, 10, 20, 50, 200-day)
  • Relative strength vs. sector and market benchmarks
  • NLP sentiment scoring on news headlines
  • Chart pattern recognition via rule-based + ML hybrid models

Stage 3: Scoring Engine (45–75 seconds)

The 127 features feed into our 5-pillar weighted scoring model. Each pillar produces a sub-score (0–100), which are combined using dynamically adjusted weights based on the current market regime (trending, range-bound, or volatile).

The output: one Confidence Score per stock, updated every trading session.

Stage 4: Delivery & Alerting (75–90 seconds)

Scores are pushed to our API, cached at the edge via Vercel, and delivered to users. The alert engine simultaneously scans for:

  • Score band changes (e.g., a stock moving from Hold to Buy)
  • Golden crosses and death crosses
  • Volume surge anomalies (>2x 20-day average)
  • Bulk/block deal detections
  • Earnings surprise triggers

Infrastructure Stack

  • Compute: Python-based scoring engine on AWS Lambda (serverless, auto-scaling)
  • Database: PostgreSQL (TimescaleDB extension for time-series) + Redis for hot caches
  • API: FastAPI with async endpoints, <50ms p99 latency
  • Frontend: Next.js 15 on Vercel Edge — ISR for scores, SSR for stock detail pages
  • ML pipeline: scikit-learn + XGBoost for pattern recognition; PyTorch for experimental sentiment models
  • Monitoring: Grafana + Prometheus, PagerDuty for on-call

Why Speed Matters

In equity markets, stale data is dangerous data. A score computed 2 hours ago doesn't reflect a breaking news event or a sudden FII sell-off. Our 90-second end-to-end pipeline ensures that when you open Stoxmine, you're seeing the most current intelligence available — not yesterday's analysis.

We're working to bring this down to under 30 seconds with streaming ingestion — stay tuned.

Ready to see Stoxmine in action?

Join 12,000+ investors who get institutional-grade intelligence — before the crowd.