// Navigation

Jacksonville Beach, FL

// Project Portfolio

Industrial-Grade
Engineering.

We take on the projects that other agencies cannot handle. From defense robotics to enterprise AI—built right here in Duval County.

sbir_phase_ii.mp4
Project 01

SBIR Phase II

Military Robotics AI Vision

The Mission: Operating at the bleeding edge of defense technology, Duval Software is actively executing a Phase II SBIR contract to integrate advanced AI Vision capabilities into military training robotics.

The Build: Our engineering team is building proprietary computer vision models that allow physical robots to process, react, and adapt to their environments in real-time. We are engineering autonomous, high-response hardware designed to elevate the training standards of the United States military.

The Client: Built in partnership with BG Workforce. Visit their website to learn more about their defense workforce and SBIR program partnerships.

PythonComputer VisionEdge ComputingNeural Networks
Project 02

The AI Battle Station

Extreme-Performance Local Compute

The Mission: Born from our roots running Overkill Gaming, the AI Battle Station represents our mastery over extreme-performance hardware. We don't just write code; we build the physical fortresses required to run it locally at maximum capacity.

The Build: We engineered a localized, hyper-powered compute station designed specifically to run massive, high-parameter AI models entirely offline. A zero-latency environment that gives elite developers and researchers absolute control over their data and models, free from cloud-based bottlenecks.

The Client: Top Secret — client identity withheld under contract terms.

Custom HardwareLocalized LLMsGPU OptimizationDeep Learning
battle_station.jpg
The AI Battle Station — a massive, custom-built computing rig with multiple monitors in a dark, high-tech workspace, engineered by Duval Software for running large language models locally.
vetxpert_architecture.png
VetXpert system architecture diagram — a Rails 8 monolith with Hotwire/Turbo/Stimulus/TailwindCSS frontend, Devise + SuperAuth with MFA/TOTP, ActionCable real-time channels, 59 ActiveRecord models with AES-256 encryption, GoodJob background workers, PostgreSQL 17 with encrypted PHI, ClamAV virus scanning, Kamal/Docker zero-downtime deploys, and integrations with OpenAI, Stripe, Twilio, Postmark, and DigitalOcean Spaces. HIPAA-compliant with full audit trail.
Project 03

VetXpert

Rails 8 Monolith Enterprise Platform

The Mission: Enterprise-grade software requires zero compromises when it comes to security, scale, and speed. VetXpert was brought to Duval Software to architect a HIPAA-compliant B2B platform serving veterans, lawyers, and admin teams handling complex operational data and protected health information.

The Build: We shipped a Rails 8 monolith with Hotwire, Turbo, Stimulus, and TailwindCSS — 59 ActiveRecord models with AES-256 encryption, Devise + SuperAuth MFA, ActionCable real-time channels, GoodJob background workers, PostgreSQL 17 with encrypted PHI, and zero-downtime Kamal deploys. 665 commits, 3,574 tests, fully audited for HIPAA compliance.

The Client: Built in partnership with VetXpert Solution. Visit their website to learn more about how the platform serves veterans, attorneys, and administrators nationwide.

Ruby on Rails 8Hotwire / TurboPostgreSQL 17HIPAAKamal / Docker
Project 04

IntelXpert

On-Premise AI Intelligence Pipeline

The Mission: Investigators handling sensitive case material can't ship evidence to a cloud LLM. IntelXpert was built to deliver large-language-model intelligence on a fully air-gapped workstation — transcription, multi-analyzer reasoning, and entity extraction with zero external network calls.

The Build: A React 18 + Vite + Zustand investigator workstation talks through a Caddy reverse proxy to a Node.js + Express orchestrator, which fans work out across 13 parallel analyzer slots. Local inference runs on a vLLM server (gemma-3-4b-awq) and Faster-Whisper + Pyannote on NVIDIA RTX PRO 5000 Blackwell GPUs. SQLite WAL persistence, file storage, and audio/PDF ingestion — all on a single offline machine.

The Client: Built in partnership with Xpert-Worx. Learn more about IntelXpert and the wider Xpert-Worx intelligence platform on their website.

vLLM / Local LLMsReact 18 + ViteNode.js + ExpressNVIDIA RTX PROAir-Gapped AI
intelxpert_architecture.png
IntelXpert system architecture diagram — an on-premise AI intelligence pipeline with a React 18 SPA + Vite + Zustand investigator workstation, Caddy reverse proxy with TLS termination, Node.js + Express orchestration with REST API and JWT middleware, WebSocket Manager for run status broadcasts, Workflow Orchestrator with file queue and 13 parallel slots, Security Analyzer Pool (CrypticLanguage, CriminalActivity, Contraband, AssaultPlanning, ThirdPartyCrime), Detective Analyzer Pool (CriminalNetworks, EvidenceWitness), local AI services on vLLM Inference Server with gemma-3-4b-awq and Faster-Whisper + Pyannote on NVIDIA RTX PRO 5000 Blackwell GPUs, and persistence via SQLite WAL and file storage.

// Ready to Build?

Your project could be next.

We only partner with visionary founders who have the capital to do it right and a project that pushes boundaries.