Work / 01

Projects &
professional experience.

A curated set of work spanning healthcare AI, multi-agent systems, generative models, and applied research. From startups I co-founded to research labs I contributed to.

01 / Experience

Roshan AI

February 2026 to Present

FounderTucson, Arizona

Developing ShifaMind, a novel interpretable clinical AI system outperforming frontier and baseline models on ICD-10 coding, incorporating a Multiplicative Concept Bottleneck (MCB) architecture with generative design extensions.
Leading end-to-end research and productionization of clinical NLP systems; coordinating with academic collaborators and preparing research outputs for top-tier venues.

Quelea Social

October 2024 to Present

Co-Founder / Lead AI EngineerSingapore (Remote)

Designed and trained a multimodal multi-agent video understanding system using OpenAI agents, OpenCV, and HuggingFace Transformers; built a NexTGPT-inspired model trained on 5k+ influencer videos achieving SOTA performance, leveraging a mixture of experts (MoE) for model scaling; prepared for arXiv publication.
Fine-tuned and deployed Qwen 2.5 Vision on 7k data points using QLoRA and self-supervised learning on RunPod, significantly improving overlay text extraction accuracy in production pipelines.

University of Arizona

January 2025 to February 2026

Graduate Research AssistantTucson, AZ

Designed and architected end-to-end ML pipelines integrating RAG, multi-agent LLM systems, and knowledge graphs across the full data lifecycle — ingestion, preprocessing, training, and cloud deployment on AWS.
Developed digital phenotyping pipelines integrating MindLamp and deploying RADAR-Base on AWS EC2 for continuous ingestion of sensor, behavioral, and survey data; built knowledge graphs, KNN clusters, and clinical text embeddings.
Engineered a multi-agent architecture unifying wearable data, passive smartphone signals, clinical text, and survey streams into a real-time analytics framework.

IonTheFold

June 2025 to Present

Co-Founder / AI Engineering LeadSingapore (Remote)

Co-founded a research-focused startup reimagining protein and antibody design with AI, collaborating with Grinter Lab (University of Melbourne) and Knott Lab (Monash University).
Designed a fusion architecture combining ESM-2, APBS, and ProteinMPNN with a novel electrostatic-aware loss function, achieving 6.53% improvement on charged proteins and 4.14% ± 0.026 overall gain in sequence recovery.

Lumenci

January 2024 to December 2024

Software EngineerGurgaon, India

Architected and deployed an automated ML-driven portfolio analysis system, reducing consultant effort from 40–50 hours per project to minutes; engineered feature extraction workflows and scalable data pipelines.
Accelerated large-scale portfolio analytics by 82% using GCP GPU infrastructure; built a multi-agent AI workbench integrating a mixture-of-experts (MoE) architecture, improving efficiency by 70% and boosting model accuracy from 6% to 72%.

Indian Institute of Technology, Indore

January 2023 to October 2023

Research InternIndore, India

Developed a Graph Neural Network architecture from scratch to process and train on a decade's worth of image data from NASA's SOHO satellite, achieving 91% accuracy in time series prediction.
Led a team of 3 researchers experimenting with CNN, RNN, GAN, and GNN for time series forecasting and image classification tasks.

02 / Projects (15)