PRODUCT R&D ROADMAP · ASSISTIVE AI & ON-PREMISES COMPUTE

What We're Building

Sparks Digital LLC builds AI and automation that help people with disabilities live, learn, and work more independently. Alongside our public-document accessibility software for local governments, we are engineering local-first AI systems that run demanding models directly on dedicated hardware.

Flagship Platform · In Development

Adaptive Learning Engine

An AI learning-support platform that builds an individualized learning model for each person and adapts pacing, format, reminders, and practice to how they actually learn.

Standard educational and workplace software forces every learner through the same rigid sequence of pages and timers. Our adaptive learning engine is designed first as practical learning support for people with ADHD and other learning differences. It observes where attention drops, which explanation formats click fastest, and when micro-step prompts help a learner finish complex work.

Why it requires substantial local computing power: Maintaining a private, low-latency personal learning model while running multi-modal reasoning and speech synthesis locally requires large GPU memory, large system RAM, and fast storage. Running on the learner's own hardware keeps personal study patterns completely private.

Scope Notice: The Adaptive Learning Engine is an educational and productivity learning-support software tool only; it is not a medical device, clinical assessment, or healthcare service, and it does not replace a licensed professional.

ACTIVE ENGINEERING & PLANNED SYSTEMS

Local-First Assistive & Municipal R&D

Every project below is labeled with its current development stage.

In Development

Private On-Site AI for Local Government

Runs our PDF accessibility scanner, optical character recognition (OCR), and document AI fully offline inside municipal buildings. Public records and internal drafts never leave the building or go to a third-party cloud.

Compute profile: Running vision-language layout models, dense OCR, and structural tag inference completely offline requires large GPU memory, large system RAM, and fast local storage.

In Development

Multi-Agent Automation

Many AI agents working in parallel to scan, fix, check, and report across large document collections. This architecture turns multi-year municipal PDF backlogs into organized, human-reviewable queues.

Compute profile: Running multiple concurrent document-analysis and verification agents simultaneously requires substantial RAM capacity, large GPU memory, and fast scratch storage.

In Development

Kindwarden On-Device Voice AI

Real-time call screening and family code-word checks to help protect older adults and people with disabilities from phone scams, with the voice models running locally.

Compute profile: Sub-second audio transcription, speaker-intent analysis, and voice synthesis without cloud round-trips rely on dedicated GPU memory and fast local storage.

In Development

IronBreach

A fully procedural 3D game built entirely from code-generated geometry, used as an engineering testbed for real-time graphics and AI agents.

Compute profile: Real-time procedural mesh generation, physics kinematics, and simultaneous multi-agent simulation demand sustained GPU memory and high-bandwidth system RAM.

Planned

Field Demonstrations

Running the full scanner and AI stack live and offline on a laptop at town halls, libraries, and schools across Indiana.

Compute profile: Executing the full offline document-remediation and vision stack during live on-site demonstrations requires mobile workstation GPU memory, large RAM, and fast NVMe storage.

Live

Indiana Property & Parcel Data API

A programmatic REST and OpenAPI data service providing normalized Indiana parcel, assessment, and municipal boundary lookups for software developers and civic researchers.

Status: Live edge API maintained by Sparks Digital LLC.