Veda BYTES: From Fragmented Crime Records to Connected Intelligence
Engineering an AI-Powered Criminal Network Analysis System
Introduction
Modern criminal investigations generate enormous amounts of information. Case records, reports, scanned documents, financial information, government records, locations, organizations and other sources may contain important clues about an investigation.
The challenge is not simply collecting this information.
The real challenge is connecting it.
A person mentioned in one document may appear under a slightly different name in another. An organization may be connected to multiple individuals and cases. A location may occur across several investigations. A financial record may provide a relationship that is not obvious from the original case document.
When these connections remain hidden inside separate records, investigators have to manually search across multiple sources to reconstruct the bigger picture.
This is the problem that our project, Veda BYTES, aims to address.
Veda BYTES is an AI-powered Criminal Network Analysis System designed to transform fragmented crime information into connected intelligence using artificial intelligence, knowledge graphs, semantic search and evidence-grounded analysis.
The core concept is to move from:
Searching individual records to: Understanding relationships between records.
The project proposes a unified intelligence platform capable of integrating structured and unstructured information and representing relationships through a knowledge graph.
The Problem
Traditional investigation workflows often require information to be examined source by source.
Consider a simplified example.
An investigation contains:
Person: Rahul Kumar
Case: Property Theft
Location: ChennaiAnother document contains:Person: Rahul K.
Organization: XYZ LogisticsA third record contains:
Organization: XYZ Logistics
Vehicle: TN-XX-XXXXA fourth record contains:
Vehicle: TN-XX-XXXX
Location: BengaluruIndividually, these records may not reveal much.
But when connected:
Rahul Kumar
│
↓
XYZ Logistics
│
↓
Vehicle
│
↓
Bengaluruthe investigation becomes a network rather than a collection of isolated records.
This relationship-first approach is one of the central ideas behind Veda BYTES.
The Veda BYTES Approach
The proposed architecture consists of several interconnected layers.
Data Sources
↓
Data Ingestion
↓
OCR / NLP
↓
Entity Extraction
↓
Entity Resolution
↓
Knowledge Graph + Vector Search
↓
Intelligence Engine
↓
FastAPI
↓
React Investigation DashboardThe proposed technology stack includes React, FastAPI, PostgreSQL, Neo4j, Qdrant, Tesseract and Python.
1. Data Ingestion
The system is designed to accept information from multiple sources.
These may include:
- Police records
- Case documents
- Scanned documents
- Financial records
- Government records
- Open-source information
- Structured databases
The first step is to bring these different sources into a unified processing pipeline.
2. OCR and Document Processing
A large amount of investigation information may exist in scanned documents.
Get Ankitkumarjhajharia’s stories in your inbox
Join Medium for free to get updates from this writer.
Therefore, Veda BYTES includes an OCR processing layer.
Scanned Document
↓
OCR
↓
Extracted Text
↓
NLP ProcessingTesseract is part of the proposed technology stack for OCR processing.
3. Entity Extraction
Once text has been extracted, the system identifies important entities.
For example:
"Rahul Kumar met representatives
of ABC Finance in Chennai."
↓
Person → Rahul Kumar
Organization → ABC Finance
Location → ChennaiThese entities become building blocks for the intelligence graph.
4. Entity Resolution
Entity extraction alone is not sufficient.
The same entity may appear in multiple forms:
Rahul Kumar
Rahul K.
R. Kumar
RAHUL KUMARThe system therefore needs an entity-resolution layer to determine whether different references potentially represent the same underlying entity.
This is important because incorrect entity matching can create incorrect relationships.
5. Knowledge Graph
The extracted entities and relationships are represented as a graph.
For example:
┌───────────────┐
│ Rahul Kumar │
└───────┬───────┘
│
ASSOCIATED_WITH
│
↓
┌───────────────┐
│ XYZ Logistics │
└───────┬───────┘
│
INVOLVED_IN
│
↓
┌───────────────┐
│ Case 1024 │
└───────────────┘Neo4j is proposed as the graph database for this layer.
Hybrid Intelligence
Veda BYTES does not depend on only one type of search.
It combines:
Graph intelligence
with
Semantic intelligence.
Graph intelligence answers questions such as: Which entities are connected to this person?
Semantic intelligence helps answer questions such as: Find documents related to financial activity involving connected organizations.
Qdrant and embeddings are proposed for semantic retrieval.
Investigator Workflow
The interface is designed around a simple workflow:
SEARCH
↓
CONNECT
↓
ANALYZE
↓
VERIFYThe investigator can search for an entity or case, explore its relationships, analyze the resulting network and finally verify the supporting evidence.
This human-in-the-loop approach is important. The system is intended to provide investigative leads, while final verification remains with authorized investigators.
Conclusion
Veda BYTES is built around a simple idea: Crime intelligence becomes more useful when isolated information is transformed into connected information.
By combining OCR, NLP, entity resolution, knowledge graphs, semantic search and geographic intelligence, the system aims to provide investigators with a unified environment for exploring complex relationships.
The project moves the investigation workflow from:
Documents → Search → Manual Connectionstoward:
Data → Entities → Relationships → Intelligence → VerificationThis is the foundation on which the rest of the Veda BYTES system is built.
