Focus Projects Research Skills Contact
AI Safety Researcher & Developer

Munish
Persaud

Computer Science and Biomedical Sciences student at UCF with research experience in AI safety for generative biology and full-stack development of healthcare AI applications. Developed novel methodology achieving 28% higher efficacy than state-of-the-art approaches in bio-foundation model safety evaluation.

View Work

Focus Areas

01

AI Safety Research

Designing end-to-end evaluation workflows and testing protocols for bio-foundational AI models. Developed methodology that bypassed existing safety protocols in Evo2 with 28% higher success rate than state-of-the-art approaches.

02

Healthcare AI

Building AI-powered systems for clinical environments, from agentic administrative automation to on-device computer vision for patient assessment. Focused on HIPAA-compliant, offline-capable architectures.

03

Full-Stack Development

End-to-end application development across web, mobile, and cloud infrastructure. Experienced with React, React Native, Flutter, Swift, and GPU-accelerated inference server deployment.

Projects

Medical Administrative Intelligence Agent

MAIA

Founder & Technical Lead

Agentic AI framework automating administrative workflows for private medical practices. Validated product-market fit through 50+ physician interviews across multiple specialties. Multi-agent orchestration with Light-RAG repository and LangGraph reduces prior authorization processing from an industry average of ~20 minutes to 2 minutes per request. End-to-end voice-note-to-EHR pipeline completes full documentation in 11 minutes per encounter, outperforming the 16-minute physician average. Custom integration pipeline enables practice-specific configuration for scheduling, billing inquiries, and documentation.

Agentic AI LangGraph Light-RAG Healthcare
Senior Design Project — UCF College of Medicine

AI-PEER

Project Manager & AI Engineer

Mobile application combining computer vision and a small language model to assess patient walking ability and reduce fall risk using the PEER clinical framework. Computer vision model processes movement analysis at 30 FPS on-device for real-time gait assessment without cloud dependency. On-device language model delivers psychological reinforcement responses in 43 seconds with zero network latency for offline clinical usability. HIPAA-compliant backend on Google Cloud Services with React Native frontend optimized for geriatric patient accessibility.

Pose Estimation Model Quantization React Native GCP HIPAA
Plastic Surgery Facial Augmentation Tool

AURA

Founder & Full Stack Developer

End-to-end application using React (web), Swift (iOS), and a fine-tuned DreamOmni2 model with LoRA to generate estimated post-surgery facial outcomes for plastic surgery consultations. GPU-accelerated inference server processes patient photos from the mobile app and returns AI-generated predictions to a clinician-facing web dashboard. Safety guardrails restrict model input and output to facial imagery only, preventing misuse.

LoRA Fine-tuning Swift React GPU Inference
Lossless Image Slide Archival

LISA

Founder & Full Stack Developer

VQ-VAE-based compression system for whole slide pathology images, achieving 98.8% file size reduction (272 MB to 3.24 MB) while preserving diagnostic quality. Built in 36 hours at SWORL (Startup Weekend Orlando) as a proof of concept. Validated through interviews with 7 medical professionals who confirmed clinical need for long-term digital slide storage and sharing. React frontend with Flask API backend deployed on a cloud GPU instance (vast.ai L40S) for real-time compression and decompression.

VQ-VAE Neural Compression Flask React
Social Events App

FOMO NOMO

Frontend Mobile Developer

Cross-platform Flutter application for campus event discovery. Translated Figma designs into responsive custom widgets ensuring consistent experience across Android and iOS. Optimized feed performance through lazy loading and infinite scroll pagination. Achieved app-wide accessibility compliance through Lighthouse audits and screen reader testing.

Flutter Cross-Platform Accessibility

Research & Work

Research Assistant

January 2025 — Present
Dr. Bedi's SAFERR AI Lab, UCF
  • Lead safety testing initiative for Evo2, a 7B-parameter generative bio-foundational AI model, designing end-to-end evaluation workflows and comprehensive testing protocols
  • Deploy and optimize multiple AI models on UCF's high-performance computing cluster, configuring containerized environments for reproducible inference pipelines
  • Analyze experimental results across 100+ test scenarios to identify safety vulnerabilities, quantify performance metrics, and inform iterative protocol improvements
  • Contribute findings toward an academic publication proposing safety improvements for bio-foundation AI models

Undergraduate Research Assistant

May 2025 — August 2025
UCF Center for Research in Computer Vision REU
  • Developed a novel methodology that bypassed existing safety protocols in Evo2, achieving a 28% higher success rate than current state-of-the-art approaches
  • Designed and implemented a comprehensive evaluation framework assessing model vulnerabilities and safety gaps across generative biology use cases
  • Demonstrated critical need for enhanced safety protocols through systematic evaluation, contributing evidence for policy recommendations in generative biology

Certified Nursing Assistant

2023 — Present
AdventHealth Fish Memorial
  • Provide direct patient care including mobility assistance, vital sign monitoring, and daily living support in a hospital setting, building firsthand understanding of clinical workflows and patient needs
  • Collaborate with nursing and physician teams to document patient status and communicate changes, reinforcing attention to detail and structured communication in high-stakes environments

Skills

Languages
Python, Java, C, C++, JavaScript, SQL, Swift
Frameworks & Libraries
PyTorch, TensorFlow, Keras, React, React Native, Flutter, Node.js, Express.js, Flask, LangChain, LangGraph
ML & AI
RAG, Light-RAG, LoRA Fine-tuning, CUDA, ROCm, High-Performance Computing Clusters
Cloud & Infrastructure
Google Cloud Services, REST APIs, Linux, Git, Docker
Bioinformatics
MACS2, HOMER, UCSC Genome Browser, BLAST
Tools & Methodologies
Jira, Confluence, Agile, Claude Code

University of Central Florida

B.S. Computer Science & Biomedical Sciences

Expected May 2026
Minor in Non-Profit Management
  • Relevant Coursework: Machine Learning (CAP 5610), Algorithms for Machine Learning, Software Engineering (CEN 5016), Robot Vision, Robotic Systems, Quantitative Biological Methods, Advanced Data Structures and Algorithms, Matrix and Linear Algebra