What it is

Paqshi is a live global intelligence platform that continuously ingests, connects, and reasons across structured and unstructured data from four domains — geopolitics, corporate markets, technology, and climate — and unifies it into a single, queryable knowledge graph.

Unlike traditional intelligence tools that operate within domain silos, or LLM-based assistants that answer questions from static training data, Paqshi builds a live ontological map of the world. Entities — nations, companies, conflicts, technologies, climate events — are extracted, deduplicated, and connected through typed, semantically named relationships. The result is a graph that doesn't just store information, but understands the structural relationships between events across domains.

The platform exposes this graph through a natural language intelligence interface backed by a Recursive Language Model reasoning engine, allowing analysts to query across domains, surface non-obvious cross-domain signals, and receive structured intelligence briefs grounded in live, traceable graph evidence.

Why I'm building it

Most intelligence and research tools are fundamentally reactive — they answer what you ask, from whatever they already know. They don't connect a conflict in the Middle East to an Indian energy company's supply chain exposure, or a semiconductor export control to a defence procurement shift, unless you already know to ask the right question.

Paqshi was built to change that. The core premise is that the most valuable intelligence isn't the answer to a known question — it's the signal you didn't know to look for. By building a cross-domain ontology that continuously updates and explicitly models relationships between entities across geopolitics, markets, technology, and climate, Paqshi can surface connections that would take a team of analysts days to find manually.

The project also grew out of a conviction that India needs sovereign intelligence infrastructure. Platforms like Palantir's Gotham exist and are deeply embedded in Western government and defence apparatus. India has no equivalent. Paqshi is a prototype of what that could look like — built domestically, running on Indian infrastructure, under Indian control.

How it works(High-Level)

At a conceptual level, Paqshi operates as a staged intelligence pipeline where raw open-source data is progressively structured, connected, and made queryable through an AI reasoning layer.

• Ingestion: Domain-specific ingestors continuously pull live data from sources across geopolitics, corporate markets, technology, and climate. Each ingestor is independently scheduled and fault-tolerant against API failures.

• Structuring & Validation: Raw items pass through domain-specific arbitration pipelines that extract structured entities, validate them against an ontology schema, and assign confidence scores. Nothing enters the graph unvalidated.

• Graph Construction: Validated entities are stored in a unified knowledge graph spanning four domains and 22 entity types. A deduplication layer resolves near-identical entities across sources. A cross-domain relation engine explicitly models typed connections between entities across domain boundaries — linking conflicts to corporate exposures, technology policies to defence capabilities, climate events to supply chain disruptions.

• Reasoning Interface: Queries are handled by two modes — a fast two-call LLM workflow for rapid analysis, and a deep recursive reasoning engine for comprehensive intelligence briefs. Both ground their outputs in live graph evidence and real-time web search.

Status

Paqshi is a working prototype with a live graph of 1600+ nodes and 2100+ relations across four domains(as of March 25th, 2026). The reasoning layer produces structured intelligence briefs grounded in graph evidence and live web search. Active development is focused on expanding data sources, improving cross-domain signal quality, and hardening the platform for production deployment.

Gallery / Demo

Paqshi Dashboard(Click for demo of previous version)