Applied Scientist · Multimodal Medical AI & Agentic Systems
I build medical AI that holds up on messy, incomplete patient data. My Ph.D. research at Georgia State targets what breaks multimodal models in practice: missing modalities, imbalanced classes, and predictions that ignore how a patient changes over time. At UST, I'm a Machine Learning Engineer II in the AI Center of Excellence, leading a 21-person team that ships agentic and RAG systems across healthcare, retail and finance.
Atlanta, GA · Contact
Citation metrics: Google Scholar, Nov 2026. Paper count includes workshops.
Ph.D. dissertation, Georgia State University. Proposal defended April 2026 · expected December 2026.
Reading a medical image involves five judgments: locating a finding, grading it, weighing it against clinical context, comparing it with earlier studies, and committing to a claim. This dissertation builds methods for each stage on two ideas: every stage carries a distribution rather than a point estimate, and evidence from different sources is represented separately so that losing one source does not corrupt the others. It is evaluated across endoscopy, whole-slide histology, retinal photography, brain MRI with genomics, and chest radiography.
Where is the evidence?
Methods: DCAN
What grade? What co-occurs?
Methods: CAP-WAE · Sparse expert routing · C²A
Which source supports the reading, and what if context is missing?
Methods: DREAM · DiA-gnostic VLVAE · SC-CoMoE
How has the finding changed since the prior study?
Methods: Dist-RRG · STRIDE
Can the claim be checked against the image?
Methods: DELTA
Recent papers presented orally at peer-reviewed venues.
Gogineni, Akash; Shaik, Nagur Shareef; Mandava, Aasrith; Masood, Adnan; Ye, Dong Hye
Proceedings of the 36th IEEE International Workshop on Machine Learning for Signal Processing (MLSP 2026), 2026
Shaik, Nagur Shareef; Cherukuri, Teja Krishna; Masood, Adnan; Ye, Dong Hye
Proceedings of the 40th Annual Conference on Association for the Advancement of Artificial Intelligence (AAAI 2026), 2026
Shaik, Nagur Shareef; Park, Jeongwoo; Kim, Yeong-Jin; Jung, Jaeuk; Oh, Hyunjung; Ye, Dong Hye
Proceedings of the 36th IEEE International Workshop on Machine Learning for Signal Processing (MLSP 2026), 2026
Shaik, Nagur Shareef; Cherukuri, Teja Krishna; Ye, Dong Hye
Proceedings of the 48th International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 2026
Systems built outside the dissertation, most of them open source.
Multi-agent healthcare ecosystem where a master intent router hands off to specialist agents (diagnosis, CBT wellbeing, triage, nutrition, visualization), including IRIS, a Vision Transformer that predicts systemic risk from a mobile retinal image, and COMPASS, an Agentic RAG insurance guide. Benchmarked on mBRSET, MIMIC-CXR, DeepEyeNet and ROCO.
Stack: Python · FastAPI · React/TypeScript · Vision Transformers · Gemini · ChromaDB
Retinal screening and HCP collaboration workspace on a DINOv2-L/LoRA ensemble (0.980 patient-level AUROC for referable DR on 247 held-out mBRSET patients), with quality-gated two-eye assessment, attention maps, clinician-signed referrals, and traceable links to guideline therapy, protein targets and drug evidence.
Stack: DINOv2-L · LoRA · FastAPI · ONNX
Explainable pedestrian traffic-risk forecaster for Atlanta (Poisson GLM + LightGBM, Empirical Bayes, weather-aware scoring) with risk-aware routing and citywide maps. Reaches 0.869 ROC-AUC on the 2023 temporal holdout, with exact risk-factor decomposition and validated AI explanations.
Stack: Python · FastAPI · React · LightGBM · MapLibre
