MIRALIA
Summary
Miralia (then Golem.ai) builds explainable, frugal AI that qualifies and routes critical inbound messages for regulated industries, especially insurance. When I joined product, delivery was stuck between ambitious Shape Up cycles and production reality: late testing, interrupt driven bugs, and a console still mostly built for internal experts.
As Product Manager I owned how we shaped and shipped InboxCare: I rewired the delivery model with engineering, rebuilt the design system around the front stack, and pulled real client feedback into the product, while pushing for a clearer product vision around explicability and frugality.
Role
Product Manager
2022 - 2025
Context
The company was focusing on a single product, InboxCare, for insurance customer service flows: categorization, extraction, routing, and reply assistance. Users were mainly solution consultants and BAs at first; clients increasingly needed to see and trust the AI’s decisions. Speed mattered for ARR; so did an experience that would hold up in demos and regulated contexts.
My role was to
Run product discovery and delivery for InboxCare features. Evolve the process from rigid Shape Up cycles to continuous, safer shipping. Align design and front end around a shared system (PrimeVue). Work directly with clients on the dashboard and turn feedback into shipped modules. Challenge the roadmap toward making explicability and frugality visible in the product, not only in the pitch.
Delivery model
We started with Shape Up (fixed 6 week cycles, flexible scope). Pitches were often too thin, infra couldn’t support continuous testing, and production bugs broke every cycle. I helped shift us to a lighter model: a Slack first “fix bugs, don’t track them” practice, stronger CI/CD with engineering, and feature flags so we could ship dark and validate on prod without breaking client demos. That unlocked smaller increments from BA and sales needs, and forced a real trade off I kept surfacing: ship fast vs. finish the UX.
Design system × engineering
We rebuilt the design system to match PrimeVue: colors, icons, components identical to code, and leaned on Figma Dev Mode so handoff almost disappeared. Reusable front components cut rebuild time and made the console more coherent as we added surfaces.
Dashboard and client feedback
For several months I spoke regularly with clients while building the dashboard. Their feedback drove concrete features: multi mailbox management, richer filters, new modules, and clearer rights. As clients started using qualification, project steering, and dashboards themselves, the product stopped being “invisible AI in Salesforce/Outlook” and became something people actually lived in.
Product stance
I argued we should treat explicability and frugality as product surfaces: visual decision traces an agent can read, CO₂ and data sobriety in the UI, not only as sales claims. Same thread: lean toward a true customer service SaaS experience (not only RPA bolted into third party tools), so feedback loops and trust sit inside the product. InboxCare Light and config automation were already pointing that way.
Outcome
Faster, safer releases; a design system shared by design and code; a dashboard shaped with clients; and a written product/strategy RETEX for leadership and the incoming CTPO on vision, quality culture, and where to invest next. The company has since rebranded as Miralia around trust, frugality, and explainable AI for critical message flows.
Next project