Case studies · Results with provenance

See the work and judge it for yourself.

Each case study shows where the work came from, what Intelligrit did, what result exists, and what that result cannot establish.

Provenance: Where the work came from
Result: What can be inspected
Role: What Intelligrit actually did
Limit: What the result cannot establish

Integrity dashboard showing an Oklahoma provider-risk overview with high-risk counts, a county map, and an investigator navigation panel. — Working Integrity hackathon interface. This is product evidence from a hackathon, with no claim of federal production deployment.

Company-led hackathon build1st place · ACT-IAC 2026

Featured record · Provider-risk investigation

Move from 548 million Medicare rows to prioritized, reviewable risk signals.

A four-person team built Integrity in 13 days using public data. It scored 1.47 million providers and concentrated known validation labels near the top while keeping the score, factors, weights, peer groups, and queries open to human review.

Rows processed: 548M
Providers scored: 1.47M
Precision@100: 42%
Lift over random: 890×

What this result shows

In this hackathon, four people processed 548 million public rows and built a usable, reviewable investigation workflow in 13 days.

What remains unproven

Production performance on current authorized data, workflow fit, false-positive and false-negative controls, fairness, security, privacy, Section 508, and Government acceptance.

Examine the evidence, method, and limitations

Evidence by question

Choose the case study that answers your question.

Open engineering shows source and tests. Working Labs shows sustained use and iteration. Timeboxed R&D answers a narrow technical question without being presented as production proof.

Open engineering

Read the code, license, tests, and architectural tradeoff.

Company open-source evidence

Public code and tests

Can a compact system remove infrastructure without giving up capability?

Keep 40+ graph algorithms. Remove the 100GB service boundary.

GraphWizard is an in-process Go graph library with 97.3% test coverage and an MIT license. Intelligrit applied it in Integrity to remove Docker and a separately operated graph service.

Read the GraphWizard case study

Intelligrit's role: Intelligrit designed and implemented the library and applied it in the company-built Integrity system.
Limit of the claim: Built and used in a company hackathon system; not federal production work.

Working Labs

See systems used, tested, and changed against real constraints.

Intelligrit Labs

Usable today

Can accessibility infrastructure be shaped by sustained daily use?

Make a blind developer's daily tools respond in speech, tones, and audio cues.

SwiftMac is working macOS-native speech infrastructure for Emacspeak, with voice control, audio icons, tones, and multi-device routing shaped by real development work.

Review the SwiftMac record

Intelligrit's role: Intelligrit built the project; Robert Melton uses it in daily development work.
Limit of the claim: Labs and open-source work used in Robert Melton’s daily development; not customer delivery.

Intelligrit Labs

Usable and evolving

Can a proven interface boundary support a more portable architecture?

Carry working speech-server lessons into a testable cross-platform system.

Omnivox evolves SwiftMac's macOS lessons into Rust platform backends, a shared audio pipeline, and concurrent streams while preserving the working Emacspeak boundary.

Review the Omnivox record

Intelligrit's role: Robert Melton and Intelligrit Labs are building and testing the system.
Limit of the claim: Usable, still-evolving Labs work; not a finished customer product or federal deployment.

Timeboxed experiments

See what each test learned without mistaking it for production proof.

Intelligrit Labs

R&D prototype

Can a constrained interface be replaced without moving server authority?

Build a modern client while gameplay, saves, rules, and processes remain server-owned.

DelveGlass places a mobile-first Godot interface over WebTiles-compatible servers as a bounded test of client replacement and integration boundaries.

Review the DelveGlass experiment

Intelligrit's role: Intelligrit Labs built the client as a bounded interface experiment.
Limit of the claim: Labs R&D prototype; not customer delivery or a production deployment.

Intelligrit R&D

Demonstration

Can one engineer turn a large extraction task into a reviewable result in days?

Turn 12 million words into traceable, searchable data in two days.

One senior Intelligrit engineer directed, reviewed, and tested an AI-assisted pipeline across 807 chapters, producing an interactive map with source-linked evidence.

Review the TWI Map demonstration

Intelligrit's role: AI accelerated extraction and implementation; one named engineer remained accountable for the result.
Limit of the claim: Recreational R&D; not client work, federal production experience, or a delivery-time promise.

Where to go next

Check the company facts, the work we offer, or ways to buy it.

Capabilities covers the company and its people. Solutionsshows the work we offer. How to Buy explains how performance, contract type, and Government acceptance fit together.

View capabilities Review the buying path