📖 TVK Nethra: System Architecture & Mathematical Whitepaper

This document provides a rigorous overview of the Nethra intelligence engine, detailing the underlying architecture, data pipelines, and the mathematical framework used to project political sentiment and render strategic visualizations.

1. System Architecture & Data Pipeline

Nethra operates on a dual-track architecture: a real-time ingestion pipeline and a highly governed SQLite inference engine. Offline sources ground the AI's understanding of historical baselines.

Spam Detection & Authentication

The Spam Block Rate metric tracks the efficiency of Quality Gate nodes:

• Spam Detection: Filters out bot farm attacks, repetitive hashtag spam (>10/min), and commercial promotional noise via NLP clustering.
• Authentication: Verifies if the source account has a history of organic political discourse in Tamil, blocking fresh "astroturfing" accounts.
• Geofencing: Uses NLP Named Entity Recognition (NER) on bio locations and post content to definitively map a post to a specific AC or GCC Ward.
• AI API Costs: Relying on lightweight LLM calls costing approximately ₹0.002 per processed social media post.

2. The Mathematical Engine: MRP & Behavioral Priors

Nethra's core predictive capability relies on Multilevel Regression and Poststratification (MRP). Baseline favorability for a demographic cell c in region i is calculated via offline public data (Census + Voter Rolls), ensuring DPDP Act compliance:

Parameter Breakdown:

• α_c (Fixed Demographic Effect): Represents inherent voting traits of a specific demographic (trained on booth vote counts).
• β X_ic (Regional Modifier): Represents how much a specific region (Kongu vs. Delta) shifts behavior.
• ε_ic (Error Margin): Irreducible statistical noise in the model.

3. Geospatial Visualization Logic (The Map)

The Geospatial Command Center plots constituencies using dynamic logarithmic radius rendering. Because voter populations (V_u) range from 150,000 to 450,000, plotting linearly makes them appear identical.

4. Competitor Favorability Breakdown (Bar Chart)

Visualizes the real-time erosion or growth of party bases. Projected vote share P(Vote_P) for party P applies NLP Behavioral Psychology Modifiers (γ) directly to the offline MRP baseline:

5. The Salience Deficit Equation (Messaging Gap)

The Messaging Salience Gap chart calculates the delta between electorate issue concerns versus TVK digital output volume:

6. Time-Series Trajectory Calculus (Trend Lines)

The 6-month trajectory line charts use an Exponential Moving Average (EMA) curve bridging historical anchors to today's live NLP baseline:

7. Historical Verification Engine & Dampening Factor

To prevent digital chatter from distorting historical realities, Nethra applies a mathematical Dampening Penalty (D = 0.85) to extreme outliers and redistributes excess share back to historical incumbents in zero-sum balance: