Research

Every project starts with a real problem. We publish.

[2 published][2 under review][8 open for collaboration]

Published

Carbon Cost of Intelligence

Authors

Rashanjot Kaur, Triparna Kundu, Kathleen Marshall Park, and Eugene Pinsky

AI Energy Research

Empirical benchmarking of energy consumption and carbon emissions across frontier LLMs on 5 MMLU domains. Establishes a per-query carbon cost framework with a 4.3× energy differential between model endpoints and 12.1% cross-domain bias in energy allocation. Foundation for the CCI service integrated into the ClinicalSearch pipeline.

LLM BenchmarkingEnergyCarbon EmissionsMMLUSustainabilityAI Efficiency

PUBLISHED · Energies · MDPI · 19(3), 642 · DOI: 10.3390/en19030642 →

Operational Resilience Under Carbon Constraints: A Socio-Technical Multi-Agentic Approach to Global Supply Chains

Authors

Rashanjot Kaur, Triparna Kundu, Bhanu Sharma, Kathleen Marshall Park, and Eugene Pinsky

Optimization AI

Multi-agentic framework for supply chain optimization under carbon regulatory constraints. Introduces the CASP metric a weighted harmonic mean for evaluating operational resilience across energy transition scenarios. Models socio-technical dependencies that static optimization ignores.

Multi-AgentSupply ChainCarbon-Aware ComputeCASP MetricEnergy TransitionOperations Research

PUBLISHED · Systems · MDPI · 14(4), 374 · DOI: 10.3390/systems14040374 →

From Hard to Soft Clustering in Machine Learning: Evaluating Deep Learning Feature Extractors for Ambiguous Multi-Label Classification

Authors

Rashanjot Kaur and Eugene Pinsky

Machine Learning Research

Evaluates deep learning feature extractors ConvNeXt, ViT, DINOv2 for soft clustering in ambiguous multi-label classification. Tests 30 configurations across 190 artists and 7 art movements. Soft clustering methods (FCM, GMM) surface 10–19% multi-movement assignments that hard clustering suppresses.

Deep LearningClusteringMulti-Label ClassificationConvNeXtViTDINOv2

UNDER REVIEW · MAKE · MDPI · Q1 · IF 6.0 · WoS & Scopus

Elementary and Robust Distribution Shape Analysis via Mean Absolute Deviations and Quantile-Based Quadrature Approximations

Authors

Triparna Kundu, Rashanjot Kaur, and Eugene Pinsky

Statistical Methods

Links quantile statistics and mean absolute deviations: MAD-based shape metrics as integrals of the quantile function with clear geometry. Midpoint quadrature recovers IQR, Galton skewness, and Moore octile kurtosis; a C-Trapezoid rule cuts approximation error and supports closed-form, outlier-resilient parameter estimation (12.5% breakdown per tail).

QuantilesMADRobust StatisticsQuadratureSkewnessKurtosis

UNDER REVIEW · Journal of Experimental and Theoretical Analyses (JETA) · MDPI

AI-Assisted Rubric-Aligned Assessment for Resource-Constrained Educational Environments

Faculty-advised research — team forming

→ Core contributors included as co-authors

Academic AI

Rubric-grounded generative assessment with self-consistency verification for high student-to-teacher ratio environments. Investigates the consistency-throughput tradeoff in automated feedback pipelines at scale. Extending to OCR ingestion for handwritten submission processing.

RAGNLPLLMPrompt EngineeringSelf-ConsistencyAutomated Grading

Apply by April 30, 2026

Explore Research →

Multi-Agent Evidence Retrieval with Citation Provenance for Resource-Limited Clinical Settings

Faculty-advised research — team forming

→ Core contributors included as co-authors

Evidence Retrieval AI

Clinical evidence exists, but frontline clinicians in resource-limited settings often lack librarians, subscriptions, and time to retrieve it reliably. This thread studies coordinated multi-agent retrieval with strict source grounding so answers stay traceable to the literature.

Multi-AgentEvidence RetrievalClinical SystemsCitation IntegrityResource-Limited Settings

Apply by April 30, 2026

Explore Research →

Cost-Constrained Multi-Agent Orchestration for Business Intelligence in Emerging Markets

Faculty-advised research — team forming

→ Core contributors included as co-authors

Business Intelligence AI

Investigates whether domain-scoped specialist agents, coordinated through lightweight orchestration, can deliver decision-grade business intelligence at cost points viable for small enterprises in emerging markets where incumbent analytics tooling is out of reach.

Multi-AgentAWSMCPCost OptimizationAgent OrchestrationBusiness Intelligence

Apply by April 30, 2026

Explore Research →

AI-Mediated Market Access for Skill-Rich, Market-Constrained Artisan Communities: A Multi-Agent Platform Approach Across Craft Verticals

Faculty-advised research — team forming

→ Core contributors included as co-authors

Artisan & Livelihoods Research

AI-mediated market access for artisan and craft communities. Field research across two regional deployment targets in partnership with IISc Bangalore.

Market AccessLivelihoodsArtisan CommunitiesPlatform ResearchPartnerships

Apply by April 30, 2026

Explore Research →

Cross-Language Agent Framework Performance Benchmarking: Google ADK Go vs Python Agent Ecosystems

Faculty-advised research — team forming

→ Core contributors included as co-authors

Agent Framework Research

First systematic performance benchmark across multi-language agent development frameworks, covering latency, throughput, memory, and framework overhead under concurrent and sustained load conditions across Go and Python ecosystems.

Agent FrameworksBenchmarkingGoPythonLangGraphPerformance

Explore Research →

Multi-Agent Specialist Debate for Abstract Visual Concept Learning Across Bongard Problem Taxonomies

Faculty-advised research — team forming

→ Core contributors included as co-authors

Visual AI Research

Multi-agent specialist architecture for abstract visual concept induction. Specialist agents coordinate through structured debate to surface visual rules that no single model reliably extracts.

Multi-AgentVisual AIConcept LearningSpecialist AgentsAbstract Reasoning

Explore Research →

Offline-First Multi-Agent Emergency Response with Privacy-Preserving Local LLM Coordination

Faculty-advised research — team forming

→ Core contributors included as co-authors

Emergency AI

Offline-capable multi-agent emergency response system with coordinated victim assistance and dispatch communication agents. Operates on locally deployed small language models without cloud dependency.

Multi-AgentEmergency ResponseOffline AILocal LLMPrivacy-PreservingHumanitarian AI

Explore Research →

Multi-Specialist Retrieval for Water Science: MCP-Enabled Evidence Access Across Hydrology Domains

Faculty-advised research — team forming

→ Core contributors included as co-authors

Environmental AI

Multi-specialist retrieval architecture adapted for water and hydrology domains. Extends the ClinicalSearch evidence access framework to water science literature with Model Context Protocol integration.

RAGMulti-AgentWater ScienceHydrologyMCPEvidence Retrieval

Explore Research →

Research collaboration? rashan@bu.edu →

Have a research direction to explore?

We review proposals with our faculty advisors. If there is genuine research potential and faculty interest, it joins the open collaboration queue. Not every proposal is accepted — what we look for is a clear problem, a realistic scope, and a reason it matters.

Research at Anacodic is faculty-advised. Accepted directions are shaped together with Prof. Park or Prof. Pinsky before any team is formed.

Submit a Research Direction →

Research

Published

Carbon Cost of Intelligence

Operational Resilience Under Carbon Constraints: A Socio-Technical Multi-Agentic Approach to Global Supply Chains

Under Review

From Hard to Soft Clustering in Machine Learning: Evaluating Deep Learning Feature Extractors for Ambiguous Multi-Label Classification

Elementary and Robust Distribution Shape Analysis via Mean Absolute Deviations and Quantile-Based Quadrature Approximations

Open for Collaboration

AI-Assisted Rubric-Aligned Assessment for Resource-Constrained Educational Environments

Multi-Agent Evidence Retrieval with Citation Provenance for Resource-Limited Clinical Settings

Cost-Constrained Multi-Agent Orchestration for Business Intelligence in Emerging Markets

AI-Mediated Market Access for Skill-Rich, Market-Constrained Artisan Communities: A Multi-Agent Platform Approach Across Craft Verticals

Cross-Language Agent Framework Performance Benchmarking: Google ADK Go vs Python Agent Ecosystems

Multi-Agent Specialist Debate for Abstract Visual Concept Learning Across Bongard Problem Taxonomies

Offline-First Multi-Agent Emergency Response with Privacy-Preserving Local LLM Coordination

Multi-Specialist Retrieval for Water Science: MCP-Enabled Evidence Access Across Hydrology Domains

Have a research direction to explore?