Lekha-Jokha
Lekha-Jokha is an AI-based investigation tool that aims to empower citizens, journalists, and scholars to make sense of the discrepancy between government claims and actual available public data.
Core trick
Orchestrates asset, audit, scheme, news, and RTI agents over fragmented public datasets, outputting claimed-versus-actual discrepancies with citations.
Description
Lekha-Jokha is an AI-based investigation tool that aims to empower citizens, journalists, and scholars to make sense of the discrepancy between government claims and actual available public data.
In India, important information related to politicians, public expenditure, audit reports, and various schemes exist. But there is no single source where all such information can be found. Hence, it becomes highly difficult for any citizen to hold their government accountable based on factual data.
This system leverages the Claude Opus 4.7 agent as a reasoning engine to orchestrate multiple domain-specific agents (Assets, CAG, Schemes, News, and RTI). On receiving an input query in natural language, our system:
Extracts structured information regarding politicians’ declaration of assets
Finds out pertinent CAG audits reports
Analyses allocations and utilization of schemes
Provides news signals
Finds RTI requests to bridge data gaps
Then all this extracted information gets aggregated into a structured dossier of "Claimed vs Actual" discrepancies with proper citation.
Unlike the conventional approach of dashboards, the Lekha Jokha is an AI-powered investigator that finds discrepancies in the fragmented data available in the public space.
README excerpt
# ⚖️ Lekha-Jokha
### AI Public Accountability Engine
> **हिसाब-किताब** — *The Reckoning*
An AI system that cross-references declared assets, CAG audit findings, scheme delivery data, and news coverage to surface accountability gaps — and drafts RTI applications for what it cannot verify.
Built for the **Cerebral Valley × Anthropic Hackathon (April 2026)** using Claude Opus.
---
## 🚨 The Problem
India publishes enormous volumes of public data:
- **MyNeta / ADR** — politician asset declarations across election cycles
- **CAG reports** — Comptroller and Auditor General findings on fiscal irregularities
- **data.gov.in** — scheme allocations, beneficiary counts, expenditure records
- **Parliamentary Q&A, news, RTI responses** — scattered across dozens of portals
None of it is synthesized. A citizen, journalist, or researcher who wants to know *"does what this politician claims match what's independently verifiable?"* faces 40+ hours of manual work across 10+ portals. Most give up.
---
## 💡 The Solution
Lekha-Jokha dispatches five Claude Opus sub-agents in parallel, each owning one evidence stream. An orchestrator synthesizes findings into a **Claimed-vs-Found dossier** — every claim citation-grounded, contradictions explicitly flagged, and unanswerable questions converted into ready-to-file RTI applications.
Type a politician name or scheme query. Within 60–90 seconds, get a structured investigation report.
---
## 🧠 How It Works
```
User Query
│
▼
┌────────