Our Work
Case studies demonstrating governed AI craftsmanship.
A verification platform for professionals in regulated or confidentiality-bound fields. The orchestrator governs the process; the app governs the artifact; the only model in the loop is yours. Citations grounded by construction, an omission defense, a refuse-or-emit export gate, and a session audit trail — shipped as a local-first application, with no model inside either tool.
View Case StudyA real AI-assisted grocery workflow missed a jar of sesame seeds — and the reason was already on record. We recover the contemporaneous decision note by a read-only query, with the honest failure included: a human approving an unverified assumption at the review step. Named honestly for what it does and doesn't prove — and a zero-PHI sample you can run in one command to recover a record yourself.
View Case StudyA comic: three AIs, one trainer, and the unglamorous habit that separates the ones who get better from the ones who get hurt. The annotated companion quotes every source verbatim and links it — plus a free four-question governance starter. Check it yourself, or hand it to your AI.
View the annotated comicA grounded website assistant that answers only from our own published pages, cites what it drew on, and — the hard part — refuses when the answer isn’t there instead of making something up. Consent-based lead capture is kept separate from the answer content and tied to a privacy policy written to match exactly what it collects. Small and single-purpose by design; the restraint is the point.
View Case StudyAn AI assistant starts every session with a blank slate — a months-long project can’t afford to. How we keep long AI-assisted work coherent: a curated project memory that survives sessions, work records whose silent alteration is detectable, and a routine check that no finding quietly falls through. A practice, not a product — described honestly, bounds first, with field results from our own working records.
Read the White Paper9 errors in a guide built from 4,392 source items. A gaming build guide that revealed three categories of AI verification failure—and a methodology applicable to legal research, software documentation, and clinical evidence.
View Case StudyAn independent research compilation on AI-generated harm and the law — the consumer AI-chatbot litigation wave and the doctrine forming around it. Nine parts, 248 citations across 43 sources, every load-bearing claim click-verifiable to its source, with the evidence base graded on a seven-tier source-trust scale.
View Case StudyA neutral, primary-source survey of Rule 11, the duty of candor, and how courts are calibrating consequences when AI-assisted filings cite cases that don't exist. The 2025–2026 sanctions wave — from admonishment to $15,000 — with every case, rule, and quotation verified against the primary record. Research, not legal advice.
View Case StudyRetrospective evaluation of a real DC small business against published DMPED scoring criteria. 34 sourced citations across 15 independent sources, with every claim linked to its source for two-click verification.
View Case StudyRetrospective evaluation of a real Atlanta nonprofit against HRSA Healthy Start Initiative scoring criteria. 778 verified data points across 17 independent sources, dual-source research architecture, and every finding linked to its source.
View Case StudyCalorie calculator, restaurant meal finder, and home meal cookbook — all built on verified USDA data, all free, all private. 11,680+ menu items, 38 recipes, zero cost to users.
View Case StudyAI-directed business transformation: research-grounded pricing strategy, domain migration, evidence-based program development, and ongoing enhancement across multiple project phases.
View Case StudyA structured comparison of five home-display calendar devices for a household purchase decision. Twenty-five independent sources, nineteen citation chain entries, every load-bearing claim linked to a verifiable passage. The methodology is the same one we bring to regulatory analysis and legal-support work — applied here to a topic any reader can sanity-check in under an hour.
View Case StudyKindergarten selection with 97.6% verified data, custom jeong framework, and live decision tool. 4 hours, 17 concerns resolved.
View Case Study4-week corrective exercise program with 30 research-grounded exercises across 3 training days, integrated movement quality tracking (Clean / Grind / Miss), and local gym facility analysis. 22 research sources, including peer-reviewed journals, with three-layer citations linking every evidence claim to its source — and, where possible, to the exact passage in the original publication.
View Case StudyA returning athlete's 10-week cut-and-restore program, grounded claim-by-claim in published research and delivered through a logging app — not a static PDF. Built on a reusable single-file engine with two additive tracking features (a cardio logger and a daily-fuel panel); nineteen findings across running, nutrition, lifting, and swimming, backed by twenty cited sources; and a live demo that runs locally with the database connection switched off. The research even overrode the athlete's own favorite running cue.
View Case StudyEcho Angel Studio has produced governed research deliverables for professional services clients under confidentiality. This includes multi-article research suites with verified citations, policy analysis frameworks, benefits research and evidence compilation, and evidence synthesis operations supporting work across legal, regulatory, and policy contexts.
The portfolio items shown here represent our public-facing work. The depth and rigor of our confidential engagements is significantly greater.
Let's Talk →