2+ years designing and deploying production AI systems — from multi-agent voice platforms and RAG pipelines to clinical knowledge graphs. Turning cutting-edge research into measurable real-world outcomes.
I'm an AI Engineer at CureMD, where I build and operate a production-grade multi-agent Voice AI platform that handles over 1,000 patient calls daily across 10+ US clinics. My systems integrate real-time speech pipelines, LLM orchestration, and clinical intelligence to deliver outcomes that actually matter.
I hold a Bachelor of Engineering in Electrical Engineering from the Institute of Space Technology (graduated Top 10%), with additional specializations from IBM, DeepLearning.AI, the University of Michigan, and Harvard's Aspire Leadership Program.
My work lives at the intersection of Generative AI, backend systems engineering, and domain-specific intelligence — particularly in healthcare. I care about systems that are observable, HIPAA-compliant, and genuinely useful at scale.
LangChain, RAG pipelines, prompt engineering, LLM-as-judge, Langfuse observability
SIP/WebRTC, LiveKit, Pipecat, Deepgram STT, ElevenLabs TTS
HIPAA-compliant platforms, EHR integrations, clinical decision intelligence
Brain tumor segmentation paper submitted — novel encoder-decoder outperforming U-Net by 5%
Neo4j, entity extraction, transformer-driven graph construction from clinical notes
A curated stack built through production deployments — every tool here has been used to ship real systems at scale.
Real systems. Real metrics. Real impact in production environments.
Production-grade multi-agent voice platform at CureMD. Handles inbound/outbound patient calls via SIP and WebRTC — automating appointment scheduling, medication refills, telemedicine, and after-hours support across 10+ US clinics with 150+ optimized prompts using the ReAct framework. Features real-time speech pipelines, HIPAA-compliant data handling, and full observability with automated QA.
Transformed 50K+ unstructured provider notes into a queryable Neo4j knowledge graph using Transformers and RAG. 92% entity extraction accuracy for patient-diagnosis relationships, enabling complex clinical querying for treatment pattern analysis.
Novel encoder-decoder architecture surpassing U-Net by 5% Dice score on BRATS 2020. Processed 3D MRI scans in PyTorch with 40% GPU utilization gain and noise suppression across 10,000+ medical images. Submitted for publication.
Machine learning system to predict blood glucose levels and assess medication impact in diabetes patients. Applied Explainable AI for model transparency and deployed a scalable FastAPI web application for clinical use.
An n8n workflow powering a Voice Coach Assistant on WhatsApp that acts as a personal AI fitness companion. Users interact via text, voice notes, or video — the system understands intent using AI and responds with training guidance, workout plans, diet reminders, and daily motivation. Integrates Google Calendar for gym scheduling and Google Maps for location-based reminders.
Industry-recognized certifications from leading institutions — click any card to verify.
Whether you're scaling AI infrastructure, automating healthcare workflows, or exploring what's possible with LLMs — I'd love to connect and collaborate.