Omnimage
I led the Design & Frontend Build for an Enterprise AI Platform that Secured $20K MRR In The First Month.
The engineering team had developed a powerful hybrid ML model capable of generating production ready product imagery at 1/10th the cost of traditional photography, boasting 92% realism accuracy. However, a model is not a product. The strategic challenge was translating this raw technical capability into an intuitive, scalable platform for non technical enterprise teams. I owned that translation.
The Gap
Enterprise product teams endure months-long lead times and six figure budgets for traditional photography. While our underlying ML architecture solved the speed and cost equation, it lacked a commercial interface. I was brought in to bridge the gap between raw algorithmic power and enterprise usability, architecting the product from zero to launch.
Key Results
- $20K MRR secured in month one post-launch
- 5 enterprise retainers signed pre-launch
- Architected and shipped a full SaaS platform with a lean 5-person team
- Established a zero-handoff Figma → Tailwind → Cursor pipeline
Hats I Wore
Product Designer
Frontend Developer
Product Strategist
Timeline
3 Months
Team
5 People — Design, Dev, ML & PM
Before Figma — Define the Real Problem
With an aggressive 3-month runway, the temptation is to immediately begin shipping UI. Instead, I forced the team to step back and define the core user barrier: trust. The technology was impressive, but enterprise users do not buy technology—they buy predictable workflows.
My core mandate was to make AI generated imagery feel as reliable as a traditional photoshoot, abstracting away the ML complexity so non technical teams could operate it daily.
Strategy: Ship a Platform, Not Just a Tool
An ML model is a feature, not a business. To close enterprise contracts, I defined the product scope to encompass a complete SaaS ecosystem. By architecting authentication, order management, version history, DAM integrations, and billing from day one, we delivered a platform that integrated seamlessly into existing corporate workflows rather than disrupting them.
Execution: AI Assisted Workflow to Match the Timeline
Given the constraints of a 3-month timeline and a 5-person team, traditional design handoff processes were a luxury we couldn't afford. I established a zero-friction pipeline: designing strictly in Figma, maintaining a single source of truth, and independently executing the entire frontend build using Cursor. This eliminated communication debt and drastically accelerated our velocity.
Constraints That Shaped the Product
/ Asynchronous latency: The UI had to gracefully mask processing delays inherent to the ML API, preserving user trust.
/ Enterprise baselines: Features like audit trails, version control, and DAM integrations were treated as table stakes, not roadmap items.
/ Parallel execution: Every design decision was scoped for immediate technical feasibility to support concurrent frontend development.
Visual Direction: Go Grayscale
I established a strict, monochromatic visual language for the platform. This was a strategic product decision, not an aesthetic preference.
When non technical users are navigating a high-complexity AI workflow, extraneous color introduces cognitive noise. By utilizing grayscale UI, we ensured the platform visually receded, allowing the generated product photography to command full attention.
This decision reduced cognitive load across every screen, increased trust with enterprise stakeholders who equated restraint with sophistication, and made the product feel premium without needing a single branded color.
*This visual restraint directly contributed to the platform's 92% task success rate during initial client onboarding.
Landing page — monochrome value prop, conversion focused from pixel one
Auth — split layout, SSO ready, trust signals baked in from the first interaction
The Hard Design Problems
Two decisions that couldn't be borrowed from a design system or solved with a component library. These required thinking through problems that had no existing pattern.
Strategic Choice One
Making the Upload Feel Guided, Not Technical
The upload interface represented the point of highest user friction. Passing instructions into a 'black box' ML model inherently triggers user anxiety. To mitigate this, I introduced an instruction review layer: before processing, the system explicitly paraphrases the user's prompt back to them in plain language.
This simple confirmation loop transformed the user psychology from uncertainty to total control.
*This friction by design approach reduced prompt related errors by 75% and eliminated generation abandonment.
I implemented a strict two-column architecture, bifurcating the user's focus: visual inputs on the left, logical instructions on the right. This structural separation of concerns drastically reduced cognitive load during complex prompt generation.
Upload flow — AI instruction review, real time processing feedback
Strategic Choice Two
Solving the Many to Many Output Problem
The platform's core UX challenge was data architecture. A user uploading 10 source images, each generating 5 variations, instantly creates a 50-node output web linked to 10 distinct inputs. Displaying this Many-to-Many relationship without overwhelming the user required a novel structural paradigm.
I designed a tripartite master-detail layout: source inputs (left), filtered output variations (center), and contextual version history (right). Selecting a source node dynamically isolates its relational outputs. To support enterprise workflows, I also architected a revertible version history, enabling teams to roll back destructive edits without incurring the computational cost of re-running the pipeline.
*This master layout reduced asset retrieval time by 60% and became the most praised feature during enterprise sales demos.
Master grid, output view, version history, reprocessing & error states
Every Surface. Designed & Built.
I maintained end-to-end ownership of the product's surface area, designing and building the frontend for complex flows including order management, API documentation, DAM integrations, and billing. The goal was to deliver enterprise-grade utility without the legacy enterprise bloat.
Orders Dashboard
Order tracking with native DAM integration — built as a baseline expectation, not a feature
Integrations & API
Plug into existing DAMs, connect via API — built for teams that already have workflows, not ones that need to change them
Settings & Account
Pricing
Three tiers — Starter, Pro, Enterprise. Clear value, no ambiguity.
Results & Impact
By maintaining tight scope and leveraging AI-assisted engineering, our lean team of five successfully shipped a comprehensive enterprise SaaS—encompassing design, frontend, ML pipeline, and infrastructure—in just three months. The strategic decisions we made around product architecture and deployment velocity generated immediate market validation.
MRR in Month One
Five enterprise clients signed retainers prior to full public launch. The product's value proposition and intuitive UX closed deals without requiring extensive pilot periods.
Person Team. Full Enterprise SaaS.
We delivered what would typically require a 20+ person organization. By establishing a zero-handoff design-to-code pipeline, we collapsed traditional development timelines.
The Cost of Traditional Photography
The hybrid ML model reliably outputs production-ready assets at a fraction of standard industry budgets, making the ROI mathematically unarguable in sales conversations.
Realism Accuracy
The outputs are indistinguishable from professional photoshoots. By wrapping this technical achievement in a trustworthy UI, we enabled non-technical teams to scale their asset generation confidently.
What I'd do differently
I would have implemented a strict component testing framework earlier in the development cycle. Because we optimized heavily for speed using Cursor and Tailwind, certain edge cases in the many-to-many output views required post-launch refactoring. Establishing a Storybook library from week one would have introduced slight upfront friction, but saved significant QA cycles during our rapid scaling phase.