Gradient Descent LLC · Virginia — Washington, DC
An applied-AI think tank, venture studio, and holding company. Research at Gradient Descent Labs, executive advisory for boards and public-sector leaders, and the parent of GOVERNBOX.ai™ — the AI, risk, IT, data & privacy system of record.
Structure
Gradient Descent LLC is the master legal entity — IP holder, contract signatory, and parent of every operating line below. Each unit serves a different buyer, so you always know which door you are walking through.
Gradient Descent LLC is a Virginia limited liability company, Entity ID 11924042, organized in 2025 and in active status with the Virginia State Corporation Commission. The firm began operations in 2026.
Applied R&D engine, academic liaison, and venture studio. Doctoral benchmarks, technical publishing, and proof-of-concept agentic systems — incubated here before commercial launch.
For: researchers, academic partners, technology leadership
High-trust executive consulting: enterprise technology strategy, board advisory, multi-cloud architecture, and executive risk — independent of any software sale.
For: boards, CEOs, federal & enterprise executives
The enterprise SaaS product line — governance in a box. GOVERNBOX.ai Advisory Services provides white-glove implementation for software clients, on the product site.
For: CISOs, CIOs, compliance officers, mid-market & public sector
/labs — Gradient Descent Labs
The Labs are where the firm's thesis gets tested: doctoral research, agentic benchmarks, and purpose-built AI systems built to run in production before anything is commercialized under GOVERNBOX.ai.
Specification-Driven Verification of Machine Learning Compliance Determinations
D.Eng. praxis · George Washington University · expected Aug 2027
Doctoral research on verifying that ML systems reach compliance determinations the way their specifications say they should.
Purpose-built models outperform generalist LLMs on constrained, high-frequency tasks — the applied thesis behind every system on this page.
Cybersecurity Analysis & Documentation Agent
Repository scanning · vulnerability analysis · MITRE ATT&CK assessment
Agentic compliance automation integrating NIST 800-53 rev 5, OWASP Top 10, MITRE ATT&CK, CVEs, and CISA KEV. Developed independently and provided to the National Endowment for the Arts for agency use, 2026.
Section 508 Analysis & Documentation Agent
Accessibility compliance evaluation
Task-specific LLM agent for automated accessibility compliance analysis. Developed independently and provided to the National Endowment for the Arts for agency use, 2026.
DevtoDeployment — Multi-Agent Orchestration
2026 · Python 3.12 · LangGraph · Google Cloud
Multi-agent system on Cloud Run, Pub/Sub, and Firestore with an agentic gateway and Claude as primary LLM backend.
Case study — governing a sensitive-domain classifier
Mental Health Text Classification — BiLSTM with attention
Graduate coursework · George Washington University
A two-layer bidirectional LSTM with an additive attention mechanism over 300-dimensional GloVe embeddings, trained in two phases to classify self-reported mental-health discourse from a public research corpus. Token-level attention heatmaps render which spans drove each prediction.
It sits here because it is the class of system this firm exists to govern. Sensitive-category inference, a consequential label, and an architecture whose attention weights can be inspected but cannot honestly be read as explanation. The questions it raises are the ones facing every agency that points a classifier at people:
Academic coursework using a publicly available research dataset. Not a screening instrument, not clinically validated, not deployed, and not offered as a product or service by Gradient Descent LLC or GOVERNBOX.ai.
Applied ML portfolio
Smaller builds that demonstrate end-to-end delivery: model, interface, deployment.
Tsunami Prediction System
Scikit-Learn · Flask · Azure App Service
Decision tree classifier (GridSearchCV, gini, max_depth 10) over seismic parameters — 89% recall on tsunami detection.
GitHub README Generator
2026 · deployed utility
Small, task-specific agent for developer documentation — the purpose-built thesis, demonstrated.
Code is published at github.com/JTunnessen.
Lab Notes — the Labs journal
Essays formerly published as ExecutiveTech — now the Labs' public notebook, alongside Forbes Technology Council contributions.
Why Identity Governance Is the Key Enabler for Secure AI Innovation
Lab Notes · 2026
One Developer. One Week. One Argument.
Lab Notes · 2026
Government Technology Is Not Broken. Government Leadership Is Afraid.
Lab Notes · 2026 · anchor essay
Beyond the Pilot: The Data Architecture of Agentic Government
Lab Notes · 2026
The Agentic Control Plane: Engineering Governance for the Autonomous Enterprise
Forbes Technology Council · 2026
Escaping Pilot Purgatory: A Framework for Scaling Enterprise AI
Forbes Technology Council · 2026
Subscribe to Lab Notes — the Labs journal, formerly ExecutiveTech. Moving to notes.gradientdescent.biz.
/advisory — Gradient Descent Advisory Services
Board-level and C-suite advisory grounded in more than 25 years of federal service — five agencies, 13 of those years at the executive level, and AI systems taken from proposal to production inside government. Independent of any software sale.
AI & Agentic Governance
Governance frameworks, agentic oversight, and responsible-AI deployment for regulated and resource-constrained environments.
Enterprise Technology Strategy
Multi-cloud architecture, platform and data strategy, and modernization programs that survive contact with the enterprise.
Security & Technology Risk
Zero Trust posture, technology risk management, and compliance programs built to be audit-ready.
Board & Executive Advisory
Fractional CIO/CAIO counsel, board education, and executive workshops on AI adoption and human augmentation.
Buying GOVERNBOX and need implementation help instead? That's GOVERNBOX.ai Advisory Services, on the product site — white-glove setup, custom governance integration, and agentic oversight configuration.
/software — the product line
The self-service AI governance platform that lets a non-technical leader stand up a defensible, board-ready governance program in about an hour — then run it, all in one place.
For nonprofits, associations, public-sector bodies, and regulated SMBs priced out of $100K+ enterprise suites.
GOVERNBOX.AI™ — U.S. trademark application pending, Serial No. 50032761, filed August 5, 2026, International Class 042. GOVERNANCE IN A BOX — Serial No. 50046079. Both owned by Gradient Descent LLC.
/leadership
Founder & Principal Researcher, Gradient Descent LLC
Founder & CEO, GOVERNBOX.ai
Technology executive at the intersection of AI governance, engineering, and public-sector leadership. More than 25 years of federal service across the Marine Corps, the Army, and five federal agencies: approximately 14 years in uniform — Marine Corps Infantry Platoon Sergeant, 1995—2002, and Army Engineer Officer, 2008—2014 — and 15 years as a federal civilian, from the General Services Administration in 2011 through the National Endowment for the Arts in 2026.
Thirteen of those years were at the executive level. A career member of the Senior Executive Service, he was twice a federal Chief Information Officer, and served as Chief Information Officer, Chief AI Officer, and Chief Privacy Officer at the National Endowment for the Arts, with prior C-suite technology roles at Voice of America, USDA's Food Safety and Inspection Service, and DHS U.S. Citizenship and Immigration Services.
As Chief Technology Officer of USDA's Food Safety and Inspection Service — an agency of roughly $1B in annual budget — he created and co-chaired the USDA Chief Technology Officer Council, coordinating technology strategy across the department's 18 agencies and 105,000 employees, against a departmental IT portfolio then exceeding $3B annually.
D.Eng. candidate in Artificial Intelligence & Machine Learning at George Washington University (expected August 2027). Forbes Technology Council member.
Affiliations
Sponsorships
Gradient Descent LLC is a Bronze Sponsor of the International Association of Privacy Professionals and the National Artificial Intelligence Association, and supports the College of Charleston Alumni Network at the Bronze level.
Sponsorship reflects support for these organizations. It does not indicate their endorsement of Gradient Descent LLC, GOVERNBOX.ai, or any product or service. Membership and sponsor marks are shown as issued by each organization.
/contact
Start with a 20-minute call: info@gradientdescent.biz
Tell us whether you're here for the Labs, the advisory practice, or the software — and we'll make sure you're talking to the right part of the firm. Virginia · Washington, DC metro.
Gradient Descent operates as a principal-led practice, with specialist staff and contractors engaged against each engagement.