Projects | Mitchell Chatterjee

Automated Cardiovascular Arrhythmia Detection - Automated ECG

Mon, 01 Sep 2025 00:00:00 +0000

This is the official code repository accompanying the paper titled Toward Robust Automated Cardiovascular Arrhythmia Detection using Self-supervised Learning and 1-Dimensional Vision Transformers.

Abstract

This project developed a robust automated cardiovascular arrhythmia detection system using self-supervised pretraining and 1D vision transformer architectures tailored for ECG signals. The system emphasizes reliability under distribution shift and low-labeled-data regimes.

Dataset & Artifacts

Public datasets: (see repository for dataset details and preprocessing scripts)
Code:
Paper: Toward Robust Automated Cardiovascular Arrhythmia Detection using Self-supervised Learning and 1-Dimensional Vision Transformers

Methods

Self-supervised pretraining on large unlabeled ECG corpora
1D Vision Transformer fine-tuning for classification and detection
Augmentation strategies, calibration, and uncertainty estimation for safer clinical deployment

Results

Achieved state-of-the-art robustness across held-out datasets.
Demonstrated improved calibration and lower false positive rates in low-label settings.

Relevance

This work underpins my research focus on medical AI and time-series modelling, and complements my engineering work on large-scale retrieval systems by bringing rigorous evaluation and robustness-first design.

Iris AI-Powered Investigative Assistant

Wed, 01 Jan 2025 00:00:00 +0000

Project Overview

Iris is 4Sight’s AI-powered investigative assistant that helps investigators quickly uncover relevant content and patterns across datasets through a natural, conversational interface. As part of JSI’s next-generation digital intelligence platform, Iris transforms how law enforcement and intelligence agencies analyze complex investigative data.

Key Capabilities

Generative AI for Search & Discovery

Natural Language Queries: Responds to investigative questions in plain English, scanning across disparate data sources
Contextual Insights: Proactively generates context-rich insights, highlighting key entities, locations, and topics for each data item
Cross-Dataset Analysis: Identifies insights and prioritizes information across multiple data types into easily digestible summaries
Follow-up Dialogue: Seamlessly transitions into conversational experience for deeper exploration and tracing connections

Language Analytics: Automatically converts speech to actionable text, streamlines transcription and translation, identifies speakers
Vision Analytics: Analyzes vast image collections to identify objects, faces, and extract text with automated content detection
Investigative Assistance: Uses advanced generative AI large language models to provide comprehensive analysis

Technical Foundation

Core Technologies (Aligned with Experience)

Transformers & LLMs: Built on state-of-the-art large language model architectures for natural language understanding
Natural Language Processing: Advanced NLP capabilities for entity recognition, sentiment analysis, and content summarization
PyTorch & Machine Learning: Implemented using modern ML frameworks for scalable AI processing
High Performance Computing: Optimized for real-time processing and analysis of large datasets

Deployment & Security

On-Premises Architecture: All AI capabilities deployed on-premises with no data sent to third-party systems
Secure & Compliant: Built to meet current and future AI regulations, with granular access controls and complete audit logging
No Internet Required: Fully air-gapped deployment ensuring data sovereignty and security

Real-World Applications

Operational Impact

Streamlined Workflows: Automates resource-intensive manual tasks like transcription and translation
Accelerated Discovery: Quickly surfaces operationally relevant intelligence the moment it arises
Enhanced Decision-Making: Provides investigators with immediate understanding of data significance and context
Global Deployment: Active across 40+ countries supporting critical law enforcement investigations

Technical Alignment

This project leverages key technical strengths from my experience including:

Transformers & LLMs for conversational AI interfaces
Natural Language Processing for multi-language content analysis
High Performance Computing for real-time processing at scale
PyTorch and modern ML frameworks for model deployment

Learn more about JSI’s AI-powered investigative solutions: and .

Sentry AI-Powered Threat Detection System

Wed, 01 Jan 2025 00:00:00 +0000

Project Overview

Sentry is 4Sight’s always-on, AI-powered risk and threat detection tool that continuously monitors incoming and historical data for high-risk signals. As part of JSI’s expanded digital intelligence platform, Sentry transforms analysis into proactive detection, reducing noise and accelerating the insight-to-action cycle for investigative teams.

Core Capabilities

Always-On Detection & Triage

Continuous Monitoring: Monitors both incoming and historical data streams 24/7 for high-risk signals
Automatic Alerts: Surfaces anomalies and operationally relevant insights the moment they arise
No Manual Searching: Eliminates the need for manual data hunting by proactively flagging threats
Configurable Categories: Agencies can configure detection categories to align with specific investigative needs

Threat Detection Categories

Threats to Life: Immediate danger situations requiring rapid response
Narcotics References: Drug-related intelligence and trafficking indicators
Suspicious Financial Transactions: Money laundering and fraud detection
Cross-Border Crime Indicators: International criminal activity patterns

Intelligent Alerting System

Clear Explanations: All alerts include traceable explanations and supporting evidence
Contextual Intelligence: Provides clear understanding of why each alert matters
Reduced Noise: Advanced filtering to surface only operationally relevant intelligence
Faster Decision Making: Gives investigative teams valuable time to focus on critical decisions

Technical Foundation

Core Technologies (Aligned with Experience)

Time-Series Analysis: Continuous temporal pattern recognition across data streams
Machine Learning: AI-powered anomaly detection and threat classification algorithms
High Performance Computing: Real-time processing capabilities for large-scale data analysis
Event Streaming: Built for high-throughput data ingestion and processing (leveraging Kafka expertise)

Deployment Architecture

On-Premises Security: Maintains data sovereignty with secure, air-gapped deployment
Compliant AI: Designed to meet security and regulatory compliance standards
Scalable Processing: Handles vast datasets while maintaining real-time response capabilities
Integration Ready: Works seamlessly within the broader 4Sight intelligence platform

Operational Impact

Investigative Enhancement

Proactive Detection: Transforms reactive investigations into proactive threat hunting
Reduced Response Time: Accelerates detection from manual processes to automated alerts
Mission Adaptability: Flexible configuration to meet diverse investigative mission requirements
Global Deployment: Supporting law enforcement operations across 40+ countries

Workflow Optimization

Automated Triage: Reduces manual workload through intelligent threat prioritization
Noise Reduction: Filters out irrelevant data to focus on actionable intelligence
Context-Rich Alerts: Provides immediate understanding of threat significance
Seamless Integration: Works alongside other 4Sight AI capabilities like Iris and Language Analytics

Technical Alignment

This project leverages key technical strengths from my experience:

Time-Series Analysis for temporal threat pattern detection
Machine Learning for automated anomaly detection and classification
Kafka & Event Streaming for real-time data processing at scale
High Performance Computing for continuous monitoring of large datasets
A/B Testing & Evaluation for optimizing detection accuracy and reducing false positives

Real-World Applications

Sentry has been deployed to help agencies detect and respond to:

Cross-border fentanyl smuggling operations
Transnational organized crime networks
Time-sensitive threats to public safety
Complex financial crime patterns requiring immediate attention

Explore JSI’s comprehensive AI detection capabilities: and .

Distributed Leader Election in a Ring

Mon, 01 May 2023 00:00:00 +0000

This project is an implementation of the min-max and min-max+ algorithms for distributed leader election in a ring. The example pictured above is flexible, the user can specify any ring size and any number of originators.

Projects | Mitchell Chatterjee

Automated Cardiovascular Arrhythmia Detection - Automated ECG

Abstract

Dataset & Artifacts

Methods

Results

Relevance

Iris AI-Powered Investigative Assistant

Project Overview

Key Capabilities

Generative AI for Search & Discovery

Multi-Modal Processing

Technical Foundation

Core Technologies (Aligned with Experience)

Deployment & Security

Real-World Applications

Operational Impact

Technical Alignment

Sentry AI-Powered Threat Detection System

Project Overview

Core Capabilities

Always-On Detection & Triage

Threat Detection Categories

Intelligent Alerting System

Technical Foundation

Core Technologies (Aligned with Experience)

Deployment Architecture

Operational Impact

Investigative Enhancement

Workflow Optimization

Technical Alignment

Real-World Applications

Distributed Leader Election in a Ring