Amir T.

I'm Amir, a biomedical AI engineer working across genomics, physiological signal processing, and medical imaging. Most of what I build has to hold up in a real clinical context, not just perform well on a clean test set. The largest system I've built end to end is a precision health platform. Patients register, fill out an intake form, then upload their test results, blood work, microbiome panels, and raw genetic data files. From there the platform runs continuous analysis, not a one-time report, and generates personalized outputs: nutrition plans, exercise recommendations, daily routines, and medical suggestions the patient can bring to their doctor. I also built out a peptide and pathway analysis layer on top of this. For the AI side, I used an ensemble approach, running multiple models together when that improved reliability, and routing to a single model when that was a better fit for the task. On top of the patient-facing side, I built a personalized medical assistant that adapts based on user feedback, plus a layer where a physician can configure their own agent that their patients then use. The whole thing runs on a proper auth system with role-based access for patients and providers. On the genomics side specifically, I work with raw genotype files, classifying variants using ACMG guidelines and cross-referencing against ClinVar and GWAS Catalog data, then generating polygenic risk scores across a few hundred conditions. I've also built parsers for lab and microbiome reports (tests like MITOswab and GI Effects), turning them into something a clinician can actually read, in English or Persian. Outside of that platform, I've worked on EEG and ECG signal processing (QRS detection, feature extraction, clustering), including sleep apnea detection from wrist actigraphy data and Parkinson's/ADHD classification using CNN/LSTM models. I've also built CNN and YOLO-based models for medical imaging tasks like bone fracture detection and lung disease classification. I've published in IEEE and SN Computer Science, and I review submissions for Frontiers in Neuroscience. That background shapes how I approach client work too, I'd rather flag a limitation in the data or the model early than let something look good in a demo and fall apart in practice. If you're building something in precision health, genomics, biosignal analysis, medical imaging, or a clinical decision support tool, I'd be glad to talk through what you're working on.