Over the last few years, I have built products in very different domains: generative AI, job search, education, and clinic operations.
Although the interfaces and users are different, several engineering lessons appeared repeatedly.
1. Multiple AI features still need to feel like one product
Klyro brings together streamed chat, cited web search, image analysis, and image generation.
The difficult part was not simply connecting several AI services. It was creating one consistent experience around them: shared authentication, one responsive interface, predictable errors, subscriptions, billing, and a transparent credit balance.
For every operation, Klyro estimates the cost, reserves the required credits, and settles the balance according to actual usage. When an operation fails, the reserved credits are released.
Live product:
https://www.klyroai.net
2. AI should assist important decisions, not hide them
RoleMint is a job-search copilot that discovers opportunities, compares them with a candidate’s profile, generates ATS-oriented materials, and organizes applications in a visual pipeline.
The user remains responsible for reviewing and approving every generated application. That human checkpoint is important when AI-generated content represents someone professionally.
3. A learning platform starts with its data model
BrightLearn supports lessons, categories, tags, quizzes, questions, answers, attempts, scoring, and progress history.
It was built with Next.js, TypeScript, Prisma, PostgreSQL, Clerk, and AI integrations. Designing the relationships between learning content and user attempts was just as important as building the interface.
BrightLearn is open source:
https://github.com/RobertIonutF/brightlearn
Application:
https://medilearn-rose.vercel.app
4. Operational software needs traceability
FizioKinetoMaxim is a protected CRM for clinic operations. It centralizes patients, appointments, medical and administrative activity, CAS cases, exercise programs, feedback, reports, incidents, complaints, and audit history.
The application uses Next.js, TypeScript, and Convex for reactive data and real-time workflows. Role-based access and traceability were essential because the application handles sensitive operational processes.
https://fiziokineto-maxim-crm.vercel.app
Building similar products
I work on complete SaaS products, AI integrations, CRM systems, educational platforms, and operational dashboards.
If you need something similar, you can contact me through LinkedIn:
https://www.linkedin.com/in/robertionutfundulea/
You can also support my independent projects here:
https://buymeacoffee.com/robertfundulea
답글 남기기