PRODUCT · UX/UI · FULL-STACK DEVELOPMENT
Fomenta
The call for proposals has rules.
The application need not be a maze.
I developed a platform for innovation funding programs: a staged application with progress, autosave, a learning track and an AI copilot that knows the call and the stage the person is on.
- MY WORK
- Product, interface, frontend and backend
- THE PRODUCT
- Multi-program platform for innovation calls
- STATUS
- Working prototype · screens with demo data
WORKING PROTOTYPEPROGRAM ENTRY
Before asking for any data,
the platform explains the program.
THE PROJECT
Writing a proposal should not
mean fighting the system.
An innovation call combines stages, text limits, documents and review criteria. The hard part should be the idea, not working out which field it belongs to, how much is left and whether what was written so far was saved.
The platform breaks the application into stages with visible progress, saves on its own while the person writes and keeps the learning track inside the same journey, instead of sending the applicant to another system.
It is a redesign study of an existing funding platform: same information and same rules, reorganised. I designed and developed both sides, applicant and program, with the API, the database and the AI layer.
Applicant
Sign-up, an eight-stage application, progress, autosave, documents, learning track and certificate.
Program
Multi-program structure by state, with each call's data isolated from the others and the call's rules enforced on the server.
Platform
Web app, own API, relational database, an AI copilot over streaming and accessibility covered by automated tests.
THE FULL JOURNEY
- Entry
- Sign-up
- Staged application
- Copilot
- Autosave
- Learning
- Certificate
CHAPTER 01 · THE APPLICATION
Eight stages, progress in sight
and nothing lost.
The dashboard opens where the person stopped and shows what is left. Each stage carries the call's limits on the screen itself, and the text is saved while it is typed: closing the browser halfway does not cost last night's work.
- Honest progress
- The bar measures stages completed under the call's rules, not the number of fields filled in.
- Autosave
- The draft is written while typing, with a note of what has already been saved.
CHAPTER 02 · THE COPILOT
The AI suggests.
The person writes.
The copilot sits beside the form and knows which stage the person is on. It explains what the call asks for in that field, answers questions and drafts text from what is already written. The suggestion appears in the copilot panel and only enters the form when the person clicks to apply it.
- Stage as context
- The copilot receives the call and the field in focus, so it answers about that passage and not about the whole program.
- Streamed answers
- The text arrives word by word over streaming, and the conversation is stored with the application.
- Applied on purpose
- No suggestion replaces what the person wrote unless they ask for it.
CHAPTER 03 · LEARNING
Learning belongs to
the same journey.
The program requires training before submission. Instead of sending the applicant to a separate platform, the track lives inside the product, with its own progress, and the certificate comes out at the end, ready to download and present.
- Track inside
- Lessons sit next to the application, with progress counting in the same account.
- Certificate at the end
- Completion produces the document, without asking the person to look for it elsewhere.
THE PRINCIPLE
No text written by AI
enters the proposal unless the person applies it.
ENGINEERING
The decisions that
hold the product up.
01One database, many programs
States and programs are a hierarchy in the data model, and every read is scoped by program. No call sees another call's data, and the rule lives in the repository layer, not in the screen.
02The copilot answers about the stage
The call's text and the field in focus go into the context of each question, which keeps the answer on topic and cheap to generate. The conversation is persisted with the application.
03Suggesting is not writing
Applying a suggested text is an explicit user action. The backend never writes into a field straight from the AI response.
04Accessibility under test
The interface has automated accessibility checks in the project itself, running with the test suite, instead of a manual review at the end.
- Next.js
- TypeScript
- Tailwind
- FastAPI
- SQLAlchemy
- Generative AI
- Streaming (SSE)
- Tested accessibility
Working prototype, built as a redesign study of an existing funding platform. The screens use demo data.
NEXT STEP
A long form does not have
to be a bad form.
If your operation depends on people filling in difficult information, the work looks like this one: understand the rule, show the progress, save without being asked and let the machine help without taking the pen out of anyone's hand.
Talk about a project
