This project focused on building and testing a practical AI security assessment lab for evaluating LLM defenses against prompt injection and jailbreak attacks.

I integrated Spikee by Reversec with a locally hosted cybersecurity model running through LM Studio, then added NVIDIA NeMo Guardrails to compare model behavior under three conditions: no guardrails, input filtering, and combined input/output protection.

The work included configuring the local model environment, building a custom FastAPI gateway, integrating NeMo Guardrails, troubleshooting model latency and timeout issues, creating a reusable Spikee target, and analyzing attack results using Spikee’s built-in reporting tools.

The project also explored different adversarial testing approaches, including prompt injection datasets, obfuscation, encoded attacks, Best-of-N testing, synthetic canary leakage tests, and structured benchmark comparisons.

The objective was to measure how much the guardrails reduced successful attacks while keeping the model, dataset, and testing conditions consistent.

Tools used: Spikee, NVIDIA NeMo Guardrails, LM Studio, Python, FastAPI, PowerShell, Parrot OS, local LLMs, JSONL datasets, and custom security testing scripts.

This project demonstrates a hands-on approach to LLM red teaming, AI safety testing, prompt-injection assessment, and guardrail validation for organizations deploying generative AI systems.

pretty cool tbh. testing the same attacks with different guardrail setups makes the results way more interesting.

AI Behavioral Health Documentation Platform

AI-Powered Clinical Documentation Built for Behavioral Healthcare Workflows

Overview

Designed an AI-powered documentation platform focused on real clinical workflows across mental health and addiction treatment. The platform transforms live conversations and uploaded transcripts into structured clinical documentation while using custom knowledge-base logic to maintain domain-specific context.

The architecture was designed around security, scalability, healthcare interoperability, and AI-assisted clinical workflows, providing a foundation for integration with existing EMR systems.

The Challenge

Behavioral health documentation requires more than basic transcription. Clinical notes need to capture nuance, context, treatment terminology, and relevant patient information while fitting naturally into existing healthcare workflows.

The platform needed to:

Convert conversations into structured clinical notes

Understand behavioral health and addiction-treatment terminology

Ground AI outputs in custom clinical knowledge

Support transcript and speech-based inputs

Maintain secure handling of sensitive healthcare data

Provide a scalable foundation for EMR interoperability

AI-Powered Documentation

The system combines speech-to-text, LLM-based summarization, clinical nuance detection, and retrieval-augmented generation (RAG) to transform raw conversations into meaningful documentation.

Custom knowledge-base logic and semantic search help ground generated content in relevant clinical information rather than relying solely on general-purpose model knowledge.

Intelligent AI Architecture

The platform uses an agent-oriented architecture combining:

AI summarization

Clinical context and nuance detection

RAG pipelines

Semantic search

Vector-based knowledge retrieval

Private LLM deployment

LLM fine-tuning capabilities

Custom healthcare knowledge bases

This creates a flexible foundation for specialized behavioral-health documentation workflows.

Healthcare-Ready Infrastructure

Security and interoperability were considered at the architecture level, including:

HIPAA-ready cloud architecture

Encryption at rest

Secure backend infrastructure

PostgreSQL + pgvector

Healthcare interoperability support

HL7/FHIR integration readiness

EMR integration support

Serverless backend architecture

Technology Stack

Frontend: React, Vite Backend: Python, FastAPI AI/LLM: AWS Bedrock, Anthropic Claude, AI Agents AI Infrastructure: RAG, Vector Database, pgvector, Semantic Search, LLM Fine-tuning Database: PostgreSQL Healthcare: HL7/FHIR, EMR Integration Cloud & Security: AWS, HIPAA-ready architecture, Encryption at Rest Speech: Speech-to-Text APIs

The Result

A scalable AI clinical documentation platform built specifically around behavioral healthcare workflows—combining conversational AI, clinical knowledge retrieval, secure infrastructure, and healthcare interoperability into a unified system.

therapists lose so many hours on notes, this solves a real problem