Rony Fozoonmehr, AI Systems Architect

Nine AI platforms · all live

Nine paradigm-setting platforms.
One creator.

Some began as private passion or commissioned projects. Others were created for the City of West Hollywood. Every one of them was dreamt up, designed, and built by me. 17 years of experienced product management, system architecture, cybersecurity, system administration, implementation experience, and technical passion were distilled into nine powerful platforms and tools that are helping shape the futures of multiple organizations.

A luminous abstract constellation representing nine AI platforms connected around civic operations — **Nine platforms, one operating philosophy.** Build the system deep enough for infrastructure, clear enough for operators, and humane enough that people actually adopt it.

First of its kind Live in city government Commercializing

Ask the city anything. Get accurate answers with verified proof and audit trails.

A true first in the world innovation. MOSAIC is the geospatial system I kept wishing cities had: a place where staff can ask deeply complex analytical questions in plain English and get back beautifully mapped results, data exports, and a clear audit trail. The magical element: the AI never sees the underlying data. It only interprets the intent. The actual answer is computed by precise mapping machinery that runs the analysis the same way every time. It's a verifiable result with a paper trail you could hand to an auditor, councilmember, or a judge. It is fully legally defensible.

MOSAIC is the geospatial system I kept wishing cities had: a place where staff can ask deeply complex analytical questions in plain English and get back beautifully mapped results, data exports, and a clear audit trail. The AI reads intent and writes structured JSON. A deterministic engine does the actual work, choosing layers, drawing buffers, and computing intersections. The AI never touches the underlying data. It translates the question; the engine proves the answer. This allows for complete transparency, full privacy, regulatory compliance, and a first in the world legally defensible system.

Dreamt up and built by me. The platform has been deployed inside the City of West Hollywood under a non-exclusive agreement, allowing the City to benefit from this transformative technology, and for me to commercialize it, using the City as a testbed and client.

Fully deterministic executionAI = intent parser onlyAI never touches dataComplete provenance per queryStructured JSON bridge

MOSAIC · Live pipeline replayIntent → JSON → Engine → Proof

AI Layer · Intent Parser

staff>

→

Deterministic Engine

Provenance trail

mosaic · GIS × permitting

MOSAIC resolving a natural-language permit query on the live city map

The result on the live map. 566 active permits within 100 ft of Santa Monica Blvd, mapped and exportable, from a single plain-English question.

Permit Pre-Review · Live replayCommercial TI · 8230 Santa Monica Blvd

Requirements validation

GIS overlay cross-reference

First of its kind In production Runs on MOSAIC

Permit Pre-Review Agent

Complete applications, before a human sees them.

Another first in the world creation. Every permit application gets a complete first review before a person ever opens the file. The agent confirms that everything required for that specific permit type is actually there, that each document is what it claims to be, and that nothing about the property raises a flag, from historic status to earthquake fault zones. It also watches the legal clock so no regulatory deadline quietly slips. Planners start from a short list of what needs attention instead of a cold read of complex applications, and reviews that used to sit for days now start clean within hours. Leveraging the MOSAIC backend is what lets it screen submissions in ways no permit screening platform in the world can.

My second first-of-its-kind system, and a personal build that shipped to production in the City. It's an orchestrator built on top of MOSAIC's deterministic geospatial pipeline: every submission is checked against the City and State's exact requirements for that specific permit type, each attachment is validated against its stated purpose, GIS overlays cross-referenced for critical overlaps like historic status, fault zones, and transit proximity, and statutory deadlines are flagged, before a planner ever opens the file. Leveraging the MOSAIC backend is what lets it screen submissions in ways no permit screening platform in the world can.

70%+faster review turnaround

100%of submissions pre-screened

Built on the MOSAIC pipelineAutomated requirements checkAttachment validationGIS overlay cross-referenceHistoric & fault zone checksStatutory deadline flagging

Production capture

agentqa · triage console

A real bug, end to end. These captures follow one live report through the console, which ships under the product name Loupe: session replayed, verdict issued, an agent re-driving the app in a real browser, and the duplicate reports already clustered. Tap a thumbnail to move through the product.

First of its kind In active use Extends any web app

AgentQA

Bug reports that arrive already reproduced.

My third first-of-its-kind platform. Most trouble tickets die before a screen share and come back as "could not reproduce." AgentQA is the solution to the most common friction point in technical customer service. When something breaks, the person clicks Report and the last few minutes of their session become a replayable Capsule, with every browser visual and console error logged. Then the platform does the part nobody has time for: a vision-enabled AI agent opens a real browser in a test environment, recreates exactly what the user did (as the user), and comes back with a verdict and the evidence. The system also helps with pre-release testing. When a vendor preps a system upgrade for release, AgentQA can replay a customer's recorded workflows to verify that they work exactly as they do on the current version. The beautiful part: it attaches to anything with a URL, which means it can extend every platform on this page and almost any SaaS product in existence, and even more incredibly, you can train it to access your test environments via the same simple recording mechanism users use to submit trouble tickets.

One shared CaptureKernel behind two surfaces: a Chrome MV3 extension for apps you don't control and an embeddable SDK whose loader stays under 5 KB and lazy-loads the rrweb capture core on first report. Redaction happens at serialization time, consent is review-before-send with no submit path until the user confirms, and Capsules upload as resumable, integrity-checked encrypted chunks into Fastify and PostgreSQL with row-level security on every tenant table. The diagnostic pipeline extracts intent and errors, resolves a safe staging environment, and reproduces the bug in real Chromium: selector replay first, then an agentic engine that stamps every interactable element with a numbered mark so a vision-language model can act by reference, verify each step against an oracle, and recover when the UI has drifted. Failures cluster into Issues by error signature and embedding, baselines diff vendor upgrades, six pluggable context strategies can reproduce as the affected user, and the AI is a local model, so session data never reaches a cloud provider.

0bytes sent without user consent

400+automated tests across the monorepo

Agentic browser reproductionSet-of-Marks vision targetingMasking at the sourceReview-before-send consentIssue clustering & deflectionUpgrade-regression baselinesLocal LLM · no cloud

haiven · live monitoring

Live monitoring. On-device detections stream in, the vehicle directory is built by edge license-plate recognition, and community alerts arrive only when actionable and with sensitive data fuzzed. Tap a thumbnail to move through the product.

In production

Community safety, with privacy as the architecture.

An autonomous community-safety platform where privacy is foundational. Facial recognition and license-plate recognition run on-device at the edge, and the data never leaves the home without the user's explicit consent. Residents use an intuitive interface to define their own alert and recording rules, and community reports are only shown when actionable and with sensitive biometric data fuzzed and fully encrypted end-to-end.

On-device facial recognitionEdge license-plate recognitionResident-defined rule engineAutomatic PII fuzzing

visionzerodashboard.weho.org

Public, live, no login. Watch the tour, or open the crash map, the yearly corridor trends, and the density heatmap. Every view runs on the city's full collision record.

Live & public City of West Hollywood

Vision Zero Analytics

Twelve years of collisions, made legible.

A public dashboard of the city's full collision record. A traffic engineer, a councilmember, or a resident can slice 2,700+ crashes by severity, mode, corridor, and time of day, watch the trends move, and see the high-injury network the way planners do. No licensing needed, no training required, completely intuitive.

0collisions mapped

12 yrsof crash data

100%public access

"You've set the standard for Vision Zero dashboards globally."

Feedback from the Vision Zero community

Open the live dashboardvisionzerodashboard.weho.org

Autonomous agent · in production

The helpdesk
that runs itself.

Triages every ticket, reads team calendars before routing, balances workload by complexity, chases deadlines, and watches sentiment. When a requester's tone shifted from calm to frustrated, it escalated on its own.

Ticket #4827 · Access issue

Agent cognitive trace

Sentiment monitor

Tone: neutralConfidence: 92%

In production City of West Hollywood

IT Helpdesk Agent

The desk runs itself.

Every incoming ticket enters an autonomous loop. The agent reads the request, checks team calendars for availability, scores the ticket's complexity, and routes it to the right person. It tracks deadlines and follows up. It also watches the conversation's tone. When frustration creeps in, it escalates before anyone asks it to.

Autonomous triage & routingCalendar-aware assignmentComplexity scoringDeadline trackingReal-time sentiment monitoringAuto-escalation on tone shift

Production capture

Autonomous agent · in production

Every contract,
against the code.

Reads every agreement against the City's contracting standards. It returns an executive summary with a risk rating, findings ranked by severity, and redlines with exact replacement language. Each finding is tied to a specific statute.

Vendor Services Agreement

AI analysis · live

Overall riskPending

In production City of West Hollywood

Contract Review Agent

A second set of eyes for counsel.

Every agreement the City signs passes through this agent first. It reads the full contract against the City's standards, produces an executive summary with an overall risk rating, ranks findings by severity, and writes redlines with exact replacement language. Each redline is cited to the municipal code or the California Civil Code. Counsel gets a marked-up document and a map of what matters, not a blank page.

Full-contract analysisRisk-rated findingsSeverity rankingRedline replacement languageMunicipal code citationsCalifornia Civil Code grounding

Production capture

playanexus · camp operations

A product tour. The member dashboard with karma and reliability scores, the live camp map, shift scheduling, and the member directory. Play the tour, or tap through the screens.

In production

PlayaNexus

Operations software for a pop-up city in the desert.

Burning Man theme camps are small cities with volunteer governments. PlayaNexus gives them a back office where non-technical leads run everything: shifts, tasks, check-ins, and the camp map. The boring parts are a game, with karma points, reliability scores, and a leaderboard that turns chores into something people volunteer for.

Full camp administrationGamified shifts & check-insKarma & reliability rankingLive spatial camp map

enclave · community management

Enclave operations view: assistance requests with agentic auto-routing, invoice audit, and a CC&R assistant answering a homeowner with citations

One operator, every association. Assistance requests auto-route and auto-assign to trusted vendors, each invoice is audited against its work order before payment, and the CC&R assistant answers homeowners with citations to the association's own governing documents and the Civil Code.

In production

Enclave

HOA management, with an AI back office.

A multi-tenant platform that runs HOA operations end to end. A retrieval assistant grounded in each association's own CC&Rs answers homeowner questions with citations, and the operational loop below runs on its own.

A request comes inA homeowner reports an issue, in plain language.

The agent triages and routesClassified, prioritized, and matched to a trusted vendor automatically.

The work order runs to completionScheduling, updates, and homeowner communication, in-platform.

The invoice is audited before paymentEvery line item checked against the work order before a dollar moves.

Rony FozoonmehrLos Angeles, CA

Seventeen years across systems administration, project management, cybersecurity, and enterprise architecture. The last twelve have been inside government, where the work has to survive audits, meetings, budgets, politics, and the person who just needs the thing to work. Add a master's in organizational psychology and four years as a union lead negotiator: I build from the metal up, and for the people layer on top.

Method

Why my systems survive
contact with real people.

Production is the only metric.

A system that isn't live is still a slide. I measure my work by what an operator can do today that they couldn't do yesterday, and by the hours that move as a result.

Depth is the advantage.

I ran the servers, the networks, and the SIEM long before I touched a model. Seventeen years of systems, security, and project delivery sit under this work, which means I do not treat AI as mist or magic. It's engineering, responsibility, and follow-through.

Governance is a feature.

I wrote a citywide AI policy with a City Attorney and built the evaluation gate every tool passes before launch. Guardrails shipped with the tools, which is exactly why people trusted them enough to adopt them.

Adoption is the finish line.

I sit with the people who own the work before I design anything, and I don't call a project done at deploy. Done is when the old spreadsheet is deleted and nobody wants it back.

"The hard part of enterprise AI isn't the model. It's the last mile into someone's Tuesday."

The record

Seventeen years of work
that made it into the world.

2009

The foundation years

Systems administration, networking, project management, and cybersecurity. The unglamorous years that make everything above trustworthy: I know what's under every abstraction I build on.

2014

Recruited to modernize a city

Brought in to rebuild West Hollywood's aging infrastructure. The role grew into organization-wide solution architecture: ERP, GIS, permitting, communications, and security.

Enterprise programs

$5.4M ERP · $1.5M permitting consolidation

Technical Lead from RFP through implementation. Three fragmented permitting systems became one, with full adoption, zero shadow systems, and departments thanking IT for the change.

2020

Zero to remote in three weeks

From no remote capability to a fully secured, organization-wide setup in under three weeks, with 100% service continuity. I also led one of California's first municipal Teams Voice deployments, cut over with zero downtime.

2020 to 2024

President, MISAC Southern Chapter

Elected by peer municipal CIOs and CTOs across Southern California to lead regional technology strategy and advise member cities on architecture decisions.

2024 to Present

A city's AI program, end to end

Nobody handed me a mandate. I wrote the strategy, co-authored the policy, built the evaluation process, shipped the city systems above, and trained the staff who use them.

Ongoing

Nine AI platforms, built end to end

MOSAIC and its permit pre-review orchestrator, AgentQA, Haiven, Enclave, PlayaNexus, the Vision Zero analytics platform, and the helpdesk and contract review agents. Multi-tenant SaaS, edge computer vision, retrieval, agentic operations, GIS, governance, and the adoption work that makes the software real. All mine, all in production.

I build AI systemspeople trust.