Capabilities · Relevant Work · Technical Approach

Artificial Intelligence
Development &
Professional Services

Zavi Digital designs, builds, and hands over production AI systems that operators own. Product engineering, multi-agent delivery, human-in-the-loop governance, and clear knowledge transfer for Cleveland Metroparks under RFP #7045.

Technical Lead Cee Zavi
Firm Zavi Digital Inc.
Focus Custom AI · Product Eng · Ops Systems
Doctrine We build it. You own it.
↓ Download PDF package Browse relevant work
01 · Technical Lead

Cee Zavi

Founder · Technical Lead · Zavi Digital Inc.

Cee Zavi leads architecture, product delivery, and AI systems engineering across Zavi engagements. Under this RFP he is the dedicated technical lead and primary counterpart for Cleveland Metroparks: use-case discovery, solution architecture, implementation oversight, security and governance reviews, and staff training.

Build

Custom AI products, ops platforms, agents, and integrations designed for real users and real traffic.

Govern

Human ship gates, data ownership, audit-friendly design, and responsible AI practices by default.

Transfer

Documentation, training, SOPs, and handoff so Metroparks staff can run and evolve the system.

02 · Alignment to Metroparks goals
  • Identify high-value AI use cases aligned to mission, guest experience, and operational efficiency
  • Design and implement scalable AI solutions with production quality, not disposable demos
  • Assess tooling and architecture; deliver roadmaps from current state to expansion
  • Train and support staff with documentation, playbooks, and adoption support
03 · Philosophy & delivery

AI that earns trust and stays owned

We treat AI as infrastructure that must earn trust: measurable outcomes, clear ownership of data and outputs, human oversight on consequential actions, and designs Metroparks can audit, retain, and export. Preferred pattern: readiness and high-value use cases, prototype for go/no-go, then harden into production with monitoring, documentation, and staff enablement.

Lifecycle · maps to RFP Scope of Work
I

Discovery & Assessment

Stakeholder needs, data hygiene, technical readiness, costs/risks, AI readiness assessment, scoped plan.

II

Solution Design

Conceptual + technical design, platform recommendations, pipelines, integrations, security alignment.

III

Development & Implementation

Models/workflows, system integration, functional/security/performance testing, pilot then production.

IV

Governance & Compliance

Risk/impact framing, explainability notes, policies/SOPs support, audit-oriented documentation.

V

Training & Support

Role-based materials, live or recorded sessions, playbooks, adoption support, incident readiness.

VI

Project Management

Dedicated technical lead / PM cadence, status reporting, SOWs for follow-on phases.

Exploratory

Discovery & learning

Prototypes, feasibility studies, readiness assessments, roadmaps. Fast learning with clear go/no-go criteria.

Outcome-driven

Production delivery

Integrated systems for a defined audience and objective: monitoring, training, and full ownership transfer.

04 · Relevant work

Production systems we have shipped

Four systems that show the work Metroparks is buying under RFP #7045: multimodal AI product engineering, full-stack product delivery with handoff, vertical operations software, and production multi-agent systems with human governance. No vapor decks. No estimated-only work presented as ours.

4
Systems detailed below
Prod
Live & launching products we actually built
HITL
Human gates on consequential AI actions
Own
Client-owned code, data, and infra paths
01 · Production multi-agent systems
In production · method + live deployments

APD/M Systems

Agentic Product Development / Management is how Zavi runs product and ops work with multi-agent teams: role-specialized agents, shared durable context, proactive observation loops, and humans who own outcomes and hard gates. It is a method and a deployable system pattern, practiced on real systems, not a chatbot pitch deck.

Minimum bar (what counts)

  • Clear product/domain and a human who owns ship / no-ship
  • Agent graph with roles (orchestrator + specialists, not one privileged chat)
  • Shared context agents can read: repo, board/issues, memory, health/metrics
  • At least one proactive loop with human-visible output
  • Explicit gates and denylists for deploy, public voice, money, destructive ops

How the system works

  • Orchestrator routes work; specialists execute (engineering, ops, content, security, etc.)
  • Shared context across git, project board, durable memory, and health collectors
  • Daily / periodic loops: collect status, cross-check commits and board, surface priorities
  • Isolation by design: separate brains/gateways, tool denylists on restricted surfaces
  • Humans own production deploy and external brand voice

Where it runs today

  • Studio product team: research, build, review, ops, delivery under human gates
  • Partner product-squad gateways: dedicated multi-agent squads for partner product work
  • External product deployments: same method bones against live external products, domain-forked context
  • Studio BD engine: research → audit → spec → prebuild → pitch with human gates on irreversible steps

Why it matters for Metroparks

Directly maps to functional and conversational agents, responsible AI, human-in-the-loop design, governance documentation, knowledge transfer, and incident-ready operations. The same pattern scales from exploratory readiness work to outcome-driven production agents with auditability and escalation.

Multi-agentShared contextHITL gates Proactive loopsIsolation

RFP map: Functional/conversational agents · responsible AI · governance · training/playbooks · production multi-agent delivery

02 · End-to-end product platform
Live · MVP engagement closed after handoff reynhome.app

Reyn

Full product engineering engagement for a homeownership OS (public brand Reyn; formerly internal Hearth). Zavi designed and shipped the MVP end to end: consumer product, marketing site, and admin operations on a shared backend, then handed the live system to the client’s founding engineering team. Engagement closed after handoff. Not an ongoing staff role.

What shipped

  • Marketing: reynhome.app - public brand and acquisition surface
  • Consumer app: my.reynhome.app - homeowner product
  • Admin ops: admin.reynhome.app - CRM, signups, engaged homes, templates, promos, operator metrics
  • Shared backend powering all three surfaces (single product system of record)

Core product domain

  • Homeowner vault and property data flows
  • Agent gift / onboarding paths into the product
  • Home health context and related operator workflows
  • Calendar / add-to-calendar and agent-detail flows late in MVP
  • Substantial app surface: multi-route Next.js product, large API layer, design system with CI discipline

Engineering depth

Production-grade engagement: modular monolith app architecture, migrations, authorization checks, test suite (including security regressions), deploy/staging discipline, and design-token hardening. Multi-agent product development was used under human direction, then formally rolled off when the client’s founding engineer took ownership of ship cadence.

Why it matters for Metroparks

Demonstrates full lifecycle: discovery and product definition, multi-surface architecture, production implementation, admin/ops tooling, documentation and knowledge transfer, and clean engagement close. Same pattern as outcome-driven park systems with staff and public surfaces.

Next.jsSupabase / PostgresMulti-surface apps Admin CRMHandoff

RFP map: Outcome-driven delivery · solution architecture · dashboards/ops · training & handoff · production AI-assisted product engineering

03 · Vertical operations software
Live product line · operator-owned stacks

BusinessOS

Productized vertical operating software for real-world operators: restaurants and booking-based service businesses. The doctrine is ownership, not platform tax: the operator keeps the software, the customer relationship, the data, and the keys. Zavi builds and hands over the spine; the business is not locked into a rented widget forever.

Restaurant vertical

  • Direct customer ordering for pickup revenue without third-party delivery app tax on curb pickup
  • Operator-controlled menu, hours, and order flow
  • Kitchen routing / back-of-house handoff for live service
  • Customer data stays with the operator, not a rented marketplace
  • Customer-facing ordering experience plus admin console as one owned system

Booking vertical

  • Quoting and scheduling spine for service businesses
  • Ops admin for jobs, availability, and field-facing workflows
  • Designed for businesses where the product is a booked service, not a menu cart
  • Same ownership model: code, infra, and data stay with the operator
  • Active build path alongside the restaurant line

Delivery model

Built once as a durable spine, then customized and handed over per operator. Deployable as client-owned infrastructure with custom domains and environment isolation. Includes the operational pattern for standing up many similar sites without becoming a multi-tenant landlord of the client’s customers.

Why it matters for Metroparks

Shows production systems thinking for staff efficiency and guest-facing service: order/request intake, admin control planes, routing of work to the right place, and data ownership. Maps to operational efficiency and service delivery goals without requiring a single consumer demo URL to make the case.

Next.jsOperator adminOrdering / quoting SchedulingOwned deploy

RFP map: Operational efficiency · service delivery · system integration · staff-facing tools · owned infrastructure

04 · Multimodal / generative AI product
Marketing live · product app in launch track rehearsalroom.app

RehearsalRoom

In-house AI product for actors and theater rehearsal. Built ground-up by Zavi: script intake and structure analysis, character/scene extraction, generative text-to-speech scene partners, speech-backed line practice, and a large public-domain catalogue plus upload path for actor material. Designed for non-technical users in real practice sessions, not lab demos.

Problem & product

Actors need reliable scene partners and structured line work without booking another human for every run. RehearsalRoom turns a raw script into a guided rehearsal environment: pick a scene, assign character voices, rehearse with audio playback, go off-book, and practice with speech recognition on the actor’s lines.

AI & multimodal pipeline

  • Document ingest: PDF / DOCX / TXT parsing into structured content
  • Chunk analysis pipeline for characters, scenes, descriptions, and lines
  • LLM-assisted script analysis for structure and character work
  • ElevenLabs TTS: voice profiles, batch generation, caching of line audio
  • Speech-to-text module for performance / line practice feedback loops

Product surfaces

  • Setup wizard: script → scene → voices → character → confirm
  • Interactive rehearsal room with playback and off-book mode
  • Scene viewer with line-level navigation and audio preload
  • Character manager: voice assignment and presentation styling
  • Catalogue: 1,350+ public-domain plays plus personal upload path

Stack & delivery

Flask backend with structured domain models (users, scripts, scenes, lines, characters, voice profiles, line audio). Modular frontend for rehearsal UX. Cloud storage for scripts and audio. Built with multi-agent production process under human ship gates. Marketing live; product app on launch track.

FlaskLLM analysisTTSSTT Supabase storageScript pipelines

RFP map: Exhibit A Project 3 (guest conversational / interactive AI) · gen AI integration · multimodal UX · non-technical audiences

05 · Exhibit A alignment

How we show up on sample projects

Exhibit A lists illustrative Metroparks AI initiatives. Below is an honest fit map, not marketing spin.

Exhibit A theme Fit How Zavi shows up
Project 1 · Zoo animal CV / analytics Partial / partner-ready Strong on data pipelines, dashboards, alerting UX, conversational query layer, multi-phase pilot→scale. CV model science is partner or specialized subtrack; Zavi leads product, integration, ops UI, and governance.
Project 2 · Park documentation audit Strong direct fit Automated evaluation of processes/docs into structured findings (gaps, strengths, recommendations). Proven in business-audit engines and report generation workflows.
Project 3 · Guest conversational AI Strong direct fit Conversational/multimodal product experience (RehearsalRoom), agent systems (APD/M), accessibility-minded UX, seasonal scale patterns from consumer products.
Honest framing: we do not claim a prior zoo computer-vision production deployment. We claim production product engineering, gen AI/agents, ops platforms, and the ability to lead architecture, integration, and governance for multi-phase AI programs.
06 · Partner skills

Mapped to RFP technical requirements

Capabilities called out under Partner Skills & Capabilities in RFP #7045.

Generative AI design & integration

Production gen AI products, agents, speech and document pipelines

API development & system integration

Edge functions, third-party APIs, auth, storage, multi-surface apps

Cloud platforms

Vercel, Supabase/Postgres, portable cloud architecture patterns

Data engineering & governance

Pipelines, RLS, ownership/export, retention-minded design

MLOps / DevOps / CI-CD

Environments, deploy discipline, staged pilot to production

UX / interaction design

Guest and staff UX, accessibility-conscious product design

Project / product leadership

Technical lead as PM/PO; SOWs, status cadence, handoff

Responsible AI / HITL

Human ship gates, escalation, monitoring-oriented design

07 · AI governance & data protection

Risk posture for public-sector AI

Because these services may process Metroparks data, the following posture informs both evaluation and any resulting agreement terms.

  • Data use: Metroparks data not used to train third-party foundation models without prior written consent
  • Ownership: Metroparks owns deliverables and outputs generated with its data; export and offboarding designed in
  • Transparency: architecture notes, known limitations, audit logs, monitoring approach per engagement
  • Human-in-the-loop: consequential actions gated; escalation for unsupported or high-risk outputs
  • Security: least privilege, secrets hygiene, Secure-by-Design defaults; NIST-oriented practices
  • Privacy & records: access controls; retention/deletion/export; public-records-sensitive handling patterns
  • Incident readiness: AI-aware plan covering privacy, security, bias/safety, model failure modes
  • Performance: measurable standards (accuracy, latency, uptime) scoped per use case with drift monitoring approach
08 · Engagement model

How work runs under the vendor pool

Under a master agreement, work proceeds project-by-project via scoped SOWs: exploratory (assessment/prototype) or outcome-driven (production delivery). Pricing models available: fixed fee for defined deliverables, blended monthly for multi-workstream programs, or T&M for tightly supervised discovery. Hosting, foundation-model API usage, and third-party SaaS remain client-owned costs unless otherwise agreed.

Ready to engage

Technical Lead: Cee Zavi · contact@zavidigital.com · https://zavidigital.com

We build production AI systems Metroparks can own, operate, and evolve.

↓ Download PDF package

Document type: Capabilities & relevant work package · Not a formal bid form set · August 2026 · RFP #7045 due Aug 11, 2026 2:00 PM EST