Case study · LensMentorA photography learning SaaS driven by an AI mentor, from RAG architecture to production
Turning an educational vision held by a non-technical expert into a differentiated SaaS product, with an AI architecture designed so that every euro spent on generation serves perceived quality.
See the live platform- Client
- LensMentor, a French photography learning platform (client: Marie-Ange Chabroullet, professional photographer)
- Role
- Full-stack developer & AI architect
The context
Marie-Ange wanted to turn her teaching expertise into a SaaS product: a platform that helps amateur photographers improve thanks to an AI mentor giving real photo critiques, tailored to each person's gear and level. The challenge was not plugging in a model but building real product differentiation: a mentor that understands the student's context.
What was done
- 1
Custom AI architecture
End-to-end design of the AI mentor, powered by Claude (Anthropic). A layered prompt-caching architecture (system, course content, camera specifications, user context, history) to optimise both the relevance of answers and the cost per request.
- 2
RAG and knowledge base
Retrieval-augmented generation over two sources (lesson content and photo critique history), indexed with pgvector on Neon and Voyage AI embeddings. A structured critique grid combining universal criteria and mode-specific criteria, across three adaptive difficulty levels.
- 3
Camera database
A database of camera bodies (Canon, Nikon, Sony, Fujifilm, Panasonic, Olympus, Pentax since 2015) that tailors advice to the actual gear: ISO, shutter speed and exposure compensation settings bounded by the camera's capabilities.
- 4
Gamification and learning paths
Seven structured paths, a unified XP system where the level declared at onboarding sets the starting point, automatically generated weekly recommendations, and anti-abuse safeguards on XP actions.
Tech stack
- Next.js
- Neon (PostgreSQL) + pgvector
- Voyage AI
- Claude API (Anthropic)
What it changes
A working, differentiated SaaS product whose AI architecture is ready to grow without generation costs exploding as usage increases.
Concrete results
First milestones approved and invoiced, AI architecture ready to be extended, product in active development.
Thibault built the LensMentor SaaS platform from scratch and the experience was excellent from start to finish. First, his responsiveness: answers come quickly and check-ins are regular, you are never left waiting. Second, he brings ideas. He doesn't just execute a specification, he challenges it, proposes relevant technical solutions and adds real value on architecture choices. Third strength: excellent development skills, with clean and reliable code. But what I appreciated most is his human side and his ability to listen. He understands what is at stake in a founder's project, adapts, and makes the collaboration smooth and pleasant. I recommend Thibault without hesitation and I keep working with him.