Active Projects

Currently In Development

Cutting-Edge AI Systems Being Built Today

5 Active Projects

Filter by Year:

Filter by Area:

In Progress

Calibrated Multimodal Radiology Copilot

2026

Clinical Decision Support & Human-in-the-Loop AI

Problem + modality: Clinical triage and reporting from medical images, radiology notes, and structured signals.
Core method: Risk-aware multimodal fusion pipeline with LLM-assisted structured outputs.
Evaluation emphasis: Uncertainty calibration and evidence-grounded output validation.

PyTorchTransformersLLM WorkflowsUncertainty Calibration

In Progress

Cross-Hospital Generalization

2026

Robust Learning Under Domain Shift

Problem + modality: Medical AI reliability across hospitals, scanners, and patient subgroups.
Core method: Standardized cross-site evaluation suite with domain-shift testing protocols.
Evaluation emphasis: Fairness slices and reproducible robustness reporting.

PythonPyTorchBenchmarkingFairness & Robustness

In Progress

Safe-to-Use Gatekeeper

2026

Trustworthy & Calibrated AI

Problem + modality: High-stakes clinical cases with uncertainty, out-of-distribution inputs, or artifact corruption.
Core method: Safety layer with abstention policies, selective prediction, and human-review deferral.
Evaluation emphasis: Coverage–risk analysis and calibration under abstention.

PyTorchOOD DetectionSelective PredictionCalibration

In Progress

Clinically Meaningful Explainability Suite

2026

Explainable Medical Image Intelligence

Problem + modality: Clinician-facing explanation needs for medical image predictions.
Core method: Concept-based explanations, counterfactual evidence, and faithfulness checks.
Evaluation emphasis: Explanation report cards and faithfulness metrics beyond heatmaps.

Grad-CAMAttention AnalysisConcept ExplanationsFaithfulness Metrics

In Progress

Evidence-Grounded LLM Assistant

2026

Vision + Language for Healthcare

Problem + modality: Structured clinical summary drafting from verified model outputs, metadata, and templates.
Core method: Safety-first LLM workflow with RAG and approved reporting templates.
Evaluation emphasis: Confidence-aware refusal and traceable citation audits.

LLM WorkflowsRAGStructured ReportingSafety Guardrails

Showing all 5 projects

Active Projects

Currently In Development

Cutting-Edge AI Systems Being Built Today

5 Active Projects

Filter by Year:

Filter by Area:

In Progress

Calibrated Multimodal Radiology Copilot

2026

Clinical Decision Support & Human-in-the-Loop AI

Problem + modality: Clinical triage and reporting from medical images, radiology notes, and structured signals.
Core method: Risk-aware multimodal fusion pipeline with LLM-assisted structured outputs.
Evaluation emphasis: Uncertainty calibration and evidence-grounded output validation.

PyTorchTransformersLLM WorkflowsUncertainty Calibration

In Progress

Cross-Hospital Generalization

2026

Robust Learning Under Domain Shift

Problem + modality: Medical AI reliability across hospitals, scanners, and patient subgroups.
Core method: Standardized cross-site evaluation suite with domain-shift testing protocols.
Evaluation emphasis: Fairness slices and reproducible robustness reporting.

PythonPyTorchBenchmarkingFairness & Robustness

In Progress

Safe-to-Use Gatekeeper

2026

Trustworthy & Calibrated AI

Problem + modality: High-stakes clinical cases with uncertainty, out-of-distribution inputs, or artifact corruption.
Core method: Safety layer with abstention policies, selective prediction, and human-review deferral.
Evaluation emphasis: Coverage–risk analysis and calibration under abstention.

PyTorchOOD DetectionSelective PredictionCalibration

In Progress

Clinically Meaningful Explainability Suite

2026

Explainable Medical Image Intelligence

Problem + modality: Clinician-facing explanation needs for medical image predictions.
Core method: Concept-based explanations, counterfactual evidence, and faithfulness checks.
Evaluation emphasis: Explanation report cards and faithfulness metrics beyond heatmaps.

Grad-CAMAttention AnalysisConcept ExplanationsFaithfulness Metrics

In Progress

Evidence-Grounded LLM Assistant

2026

Vision + Language for Healthcare

Problem + modality: Structured clinical summary drafting from verified model outputs, metadata, and templates.
Core method: Safety-first LLM workflow with RAG and approved reporting templates.
Evaluation emphasis: Confidence-aware refusal and traceable citation audits.

LLM WorkflowsRAGStructured ReportingSafety Guardrails

Showing all 5 projects