Sole Product Designer · CSGPT.ai
Voice First: Designing an AI Agent Platform from Scratch
From a simple design system to a full working product in one year. I was the only designer on a small team building CSGPT, a platform that lets companies create chat and voice AI agents in minutes, connect them to their CRM, and deploy without code.
What came out of it
1
Designer. I owned the entire product experience from design system to shipped flows.
100%
Tailored voice agent solutions aligned to each client's operational use case.
Voice + Chat
Two agent types, many flows. From reservations to HR screening to sales, each use case has its own path.
The challenge
CSGPT is a startup that helps businesses set up AI voice and chat agents in minutes, no coding required. The product needed to serve diverse industries: restaurants managing reservations, HR teams screening candidates, customer support handling enquiries, and sales teams qualifying leads. Each customer had different needs, and the product had to evolve quickly.
I joined as the sole designer in a small cross-functional team with front-end and back-end developers and a PM. We evolved an early design system into a production product within a year. Discovery sessions with companies such as Storebox and GoStudent shaped priorities around reservation management, appointment booking, and churn prevention.
I owned flow definition, component design, and interaction quality, ensuring the platform remained coherent and intuitive while we iterated quickly across multiple use cases.
What we set out to achieve
- Voice and chat agents as the two core product pillars, with flows tailored to each use case (reservations, HR, support, sales)
- No-code experience so non-technical users could set up agents in minutes
- Flexible integrations: Molzeit, Quandoo, and CRM connections, with deferral options so users could connect later
- 100% custom voice agents per client, with the platform as the engine and each solution tailored to their use case
- Design system that could scale from MVP to full product as we iterated with real customers
How we approached it
I developed the product around two pillars: voice and chat agents. Voice journeys required number setup, integration paths, and operational configuration; chat journeys needed flexible setup via conversational or prompt-based flows, file uploads, and URL scraping. The design approach balanced platform consistency with the flexibility required for different client contexts and industry-specific requirements.
Voice agents
Handle inbound and outbound calls. 100% custom per client. Phone setup, reservation integrations (Molzeit, Quandoo), and use-case-specific flows.
Chat agents
Text-based agents with conversational setup, prompt-based creation, and data onboarding via file upload or URL scraping.
Customer-driven evolution
Sessions with Storebox, GoStudent, and others shaped the roadmap. Integrations and flows came from real needs; each use case got its own path.
Voice agents
Voice agents handle phone calls for reservations, screening, support, and lead qualification. Each use case has its own flow. For restaurants, that means agent type selection, restaurant details, reservation system integration (Molzeit, Quandoo), and phone number setup. Voice agents are 100% custom per client; the platform provides the engine, not the script.
Agent type selection
Users choose from pre-configured templates: 24/7 Restaurant assistant, Marketing assistant (coming soon), Candidate screening assistant (coming soon), or Manual for full customisation. The cards use clear icons and descriptions to help users pick the right path quickly.
Restaurant information & reservation system
The form collects restaurant name, email, assistant language (25 languages supported), and opening times. The reservation system section lets users connect Molzeit or Quandoo, or defer, with a reassuring message that the assistant remains functional even without immediate integration.
Phone number setup
Voice agents need a number. We offer a dedicated Austrian number (€15/month) with 24/7 availability and call analytics, or the option to forward an existing restaurant number. Clear pricing and benefits help users decide.
Success screen
A clear completion state confirms the assistant is ready and invites users to test it. The next step is a live call to see how the agent handles reservations and customer enquiries.
Chat agents
Chat agents communicate via text. Users can describe their agent in plain language (e.g. "Make me a sales assistant that qualifies leads and books demo calls"), and the platform guides them through setup by choosing Voice or Chat, language, and data sources. File upload and URL scraping let the agent learn from company content.
Prompt-based entry
The landing experience invites users to describe their agent in a single prompt. No coding required. The platform interprets the intent and guides them through the right configuration.
Conversational setup
The AI assistant guides users step by step: name the agent, choose Voice or Chat, select language. The chat interface feels familiar and reduces cognitive load for non-technical users.
Data onboarding for chat agents
For chat agents, users upload data files (PDF, DOC, XLS, TXT) or provide a URL for web scraping. The agent learns from this content to answer questions accurately. The drag-and-drop area and clear file size limits keep the flow simple.
Key outcomes
From a lean design system to a full product in one year. Voice and chat agents each have flows tailored to their use cases, including reservations, HR screening, support, and sales, while sharing consistent UI patterns. Customer discovery directly shaped integrations and flows; voice agents remained 100% custom per client.
- Full product ownership: Sole designer from design system through to shipped flows for voice and chat agents across many use cases.
- Voice + Chat: Voice agents with phone setup, Molzeit/Quandoo integrations; chat agents with conversational setup, file upload, and URL scraping.
- Customer-driven roadmap: Sessions with Storebox, GoStudent, and others informed integrations and deferral options like "I'll connect later".
- 100% custom voice agents: Each client gets a tailored solution; the platform is the engine, not the script.
What I led and delivered
- Led product design end-to-end as the sole designer, from early system foundations to shipped voice and chat experiences.
- Used customer discovery input to prioritise features and integration decisions with clearer delivery focus.
- Balanced bespoke client needs with platform-level consistency across varied use cases.
- Established shared UI patterns that improved usability and reduced fragmentation across journeys.
- Partnered closely with product and engineering to sequence scope pragmatically under rapid iteration.