Artisan.ai

Inspiration

Creative small businesses lose leads in two ways. First, owners are busy and often do not have time to respond to every inquiry, so potential customers drop off before they ever get an answer. Second, even when they do respond, the work is still hard: customers usually ask for something vague like "modern, elegant, and photo-friendly," and the owner has to turn that into a concrete offer that feels current, matches the business's style, uses real pricing, and is actually feasible to deliver.

That means every lead becomes a time-consuming mix of sales, creative direction, market research, and quoting. We built artisan.ai to solve both problems at once: help creative businesses respond instantly to more inquiries, while turning vague customer ideas into trend-aware, owner-approved proposals.

What it does

A customer visits a business's online storefront and describes what they want in plain language. artisan.ai then:

• Understands the request by extracting the occasion, style, budget, and constraints
• Retrieves business knowledge — catalog, pricing rules, style guides, and past offerings — from a Gradient Knowledge Base using RAG
• Researches live trends so the response reflects what is popular right now, not generic or outdated ideas
• Generates three tailored proposal options — budget, mid-range, and premium — with itemized pricing
• Creates visual mockups for each option using image generation
• Routes everything to the owner for review, edits, and approval before it is shared with the customer

The result is a workflow that helps small businesses respond faster, recover more leads, and still keep the owner in control of quality and final approval.

artisan.ai is multi-tenant, so the same workflow can power many types of creative SMBs. We seeded five demo businesses — including bakeries, florists, photographers, print shops, and event entertainment — to show that the core problem is surprisingly universal: vague request to bookable proposal.

How we built it

We built this on the DigitalOcean Gradient AI Platform:

• Gradient Knowledge Bases store each business's catalog, pricing, and style information. Retrieval gives the model grounded business context instead of letting it invent prices or offerings.
• Gradient Serverless Inference powers proposal generation with openai-gpt-oss-120b through the OpenAI-compatible API.
• Gradient image generation with fal-ai/flux/schnell creates mockups for each proposal option.
• Gemini 2.5 Flash with grounded Google Search adds live trend research before the proposals are generated.

The frontend is built with Next.js 16 and Tailwind CSS 4, while SQLite stores business and proposal data. The full workflow — from customer request to tailored concepts, pricing, mockups, and owner review — is orchestrated through a single API pipeline.

Challenges we ran into

The hardest part was making the proposals feel both creative and trustworthy.

• Structured output — We needed the model to return proposal data in a clean, consistent format with itemized pricing, add-ons, and totals.
• Grounded pricing — Without strong retrieval, the model produced plausible-sounding but inaccurate offers. RAG was essential to keep the proposals tied to real business data.
• Context balance — Too little retrieved context made the outputs generic; too much made them noisy. Tuning retrieval quality was important.
• Latency — Generating three visual mockups per request adds overhead, so we parallelized image generation to keep the experience responsive.

Accomplishments that we're proud of

• We built for a real and underserved workflow. Creative SMBs do not just need help writing quotes — they need help responding faster and turning vague demand into something concrete and sellable.
• The trend-aware layer makes the proposals feel current. The system does not just generate options; it grounds them in what is happening right now.
• RAG makes the output business-specific. Each proposal reflects the actual catalog, pricing, and style of that business instead of sounding generic.
• The platform is multi-tenant from day one. The same core workflow works across several creative business types.
• The entire AI stack runs on Gradient. Retrieval, inference, and image generation all run on the same platform.

What we learned

• This is not just a quoting problem. It is a response-speed problem and an interpretation problem. Owners lose leads when they cannot reply quickly, and they lose time when every reply has to be handcrafted.
• RAG is what makes the system usable. Without business knowledge, the model is creative but unreliable. With retrieval, it becomes grounded and practical.
• Live research matters. Trend awareness makes the generated proposals feel more relevant, timely, and differentiated.
• Creative businesses share a common workflow. Whether it is a bakery, florist, or photographer, the pattern is the same: vague request → concept options → pricing → approval.

What's next for artisan.ai

• Let customers refine concepts through conversation
• Give owners more control over pricing guardrails and upsell rules
• Connect proposals directly to booking and deposits
• Add notifications for owners and customers
• Build analytics around conversion and proposal performance
• Expand to more creative business categories

Built With

• react
• vite