WesternFrontConflict intel
DashboardResourcesAbout
Live

Architecture

How the intelligence is made.

A retrieval-augmented pipeline: every item is embedded and stored, then each assessment draws on the most relevant memory across the entire history, not just the latest headlines.

Data sources

Reddit

13 subreddits

Geopolitics, regional, and country-specific communities.

r/geopoliticsr/CredibleDefenser/pakistanr/india

RSS Feeds

11 feeds

Major news outlets across South Asia.

DawnTimes of IndiaSCMPKathmandu Post

NewsAPI

Configurable

Global news aggregation, optional.

ReutersAP NewsAl JazeeraBBC

Processing pipeline

  1. 01

    Ingestion

    Collect all news from every source without filtering.

  2. 02

    Embedding

    Convert text to vectors with the local MiniLM-L6-v2 model.

  3. 03

    Storage

    Persist embeddings in ChromaDB as durable memory.

  4. 04

    Retrieval

    Query semantically relevant items via vector search.

  5. 05

    Analysis

    Generate the assessment with Gemini over retrieved context.

Vector memory

  • All news items are embedded and stored permanently
  • ChromaDB provides persistent vector storage
  • Semantic search retrieves contextually relevant items
  • Historical patterns inform the current analysis
  • No keyword filtering: vectors determine relevance

Limitations

  • Data may contain unverified information
  • AI analysis lacks human contextual judgement
  • Limited to publicly available sources
  • Potential for bias in source selection
  • Analysis quality depends on data volume

Data usage policy

All data is collected from public sources and used solely for analysis, in compliance with each platform's API terms. Embeddings are stored locally in ChromaDB. Individual items are never displayed; only the aggregated analysis is presented.