EU Funding / ERDFAI / RAGReal-time CollaborationFull Stack SaaS12-Month BuildDual Runtime

Structura — AI-Powered EU Grant Application & Project Management Platform

A full-lifecycle SaaS platform for authoring, collaborating on, and managing EU funding applications — from blank idea through structured submission to post-award project tracking. Powered by a dual-runtime architecture: Next.js for a fast localised UI and FastAPI for AI, RAG, and document-heavy processing.

12 mo

End-to-end build

Separate runtimes

Application types

RAG

AI + vector search

Real-time collab

i18n

Multi-language

The Problem

EU grants fail due to process — not project quality

Applying for EU funding like ERDF grants is one of the most complex administrative challenges organisations face. Applications are long, heavily structured, require multiple stakeholders, must comply with strict regulatory rules, and often fail simply because of poor writing, missing sections, or misunderstood criteria — not because the project itself is weak.

The client needed a platform that could guide organisations through the entire grant lifecycle — from structuring an initial idea, through collaborative application writing with AI assistance, to post-award project management, reporting, time tracking, and regulatory compliance. No such tool existed. We built it from scratch.

Architecture

Dual-runtime design

One of the core engineering decisions was splitting the system into two purpose-built runtimes — each doing what it does best.

Frontend Runtime

Next.js 15

React 19, TypeScript, Tailwind CSS 4, next-intl (i18n), NextAuth, Socket.IO client

AI & API Runtime

FastAPI

Python, LangChain, Pinecone, Anthropic + OpenAI, Socket.IO, Motor/MongoDB

Data & Storage

MongoDB

+ Pinecone vectors + AWS S3 + Resend email

Phase 1

Application Authoring

4 application types — Mini, Pre-study, Main application, ESF variant — each with its own schema-driven editor route and section layout.

Schema-driven sections — Conditional fields, repeatable groups, rich input types with validation per section.

Context Hub — AI-powered workspace per application where users upload documents and briefings that feed into all AI features.

AI writing assistant — Section-level prompts, autodraft generation, ERDF-specific writer that generates text from context and user inputs.

Streaming validation — Real-time AI validation of sections against funding criteria, cross-validation between sections, smart notes.

Export — Application data exported to formatted documents via client-side generation (docx, PDF).

Phase 2 & 3

Collaboration + Post-Award

Collaborator invites — Share links, token-based invite flows, role-aware access for internal and external stakeholders.

Threaded comments & real-time chat — Section/application-level comments and Socket.IO chat rooms with read state and unread counts.

RAG chat assistant — Document-aware AI chat that answers questions based on uploaded context and funding guidelines in real time.

Project workspace — Post-award project CRUD, dashboard aggregates, team management, and AI-assisted project import.

Weekly updates + evidence — Structured progress reports with file attachments per week, archived for audit purposes.

Time reporting & compliance — Contributor time logging with approval flows, De Minimis tracking, indicators tracking for EU reporting.

AI & RAG Features

Where the engineering depth lives

🧠 RAG Document Search

Documents vectorised via LangChain + Pinecone. All AI features search this vector store for contextually accurate, document-grounded answers.

✍️ ERDF Writer / Autodraft

AI drafts full application sections based on user context and funding guidelines. Uses Anthropic + OpenAI via LangChain with streaming output.

✅ Streaming Validation

Sections validated against ERDF funding rules in real time — streamed character by character to the UI.

💬 RAG Chat Assistant

Guided conversational assistant with full document awareness, answering questions with source citations.

📝 Smart Notes

AI extracts key insights, action items, and structured notes from application content. Saved, editable, linked to sections.

📂 Content Extraction

Uploaded files (PDF, DOCX, XLSX) parsed by PyMuPDF, pdfminer, mammoth — text extracted and fed into RAG pipelines.

Engineering Challenges

Why this was a year of work

Dual-runtime auth security

Session cookies managed in Next.js can't safely cross to a separate FastAPI origin. We built a Next.js proxy layer that derives a Bearer token from the session cookie and forwards it — keeping auth airtight without cross-site cookie risk.

Schema-driven dynamic editors

Each of 4 application types has its own section schema with conditional fields, repeatable groups, and per-field validation. Building a single editor layout that renders correctly for all types required deep schema abstraction.

Streaming AI across two runtimes

Validation and writing features stream character-by-character from FastAPI through the Next.js proxy to the browser. Managing backpressure, error states, and UI rendering across this chain was a significant engineering challenge.

Real-time collaboration at application level

Socket.IO rooms per application, coordinated between the Python backend and the Next.js frontend, with read state, unread counts, and live broadcasting — without race conditions.

RAG over complex regulatory documents

EU funding guidelines are long, dense, and full of cross-references. Building RAG that retrieves the right chunks required careful chunking strategy, metadata filtering, and prompt engineering.

Frontend Stack

Next.js App

Framework — Next.js 15.5, React 19.1, TypeScript

Styling — Tailwind CSS 4, Headless UI, Lucide

i18n — next-intl (localisation)

Auth — NextAuth + @auth/mongodb-adapter

Forms — react-hook-form + Zod validation

Real-time — Socket.IO client

Exports — docx, pdfkit, mammoth, xlsx

Storage — AWS SDK (S3 uploads + signed URLs)

Backend Stack

FastAPI AI Service

Framework — FastAPI, Pydantic, Uvicorn/Gunicorn

AI — LangChain (OpenAI + Anthropic + MongoDB)

Vector DB — Pinecone (vector search + RAG)

LLMs — Anthropic Claude + OpenAI GPT SDKs

Database — Motor + PyMongo (async MongoDB)

Real-time — python-socketio (WebSocket rooms)

Parsing — PyMuPDF, pdfminer, mammoth, python-docx

Email — boto3 (AWS S3), Resend

The Outcome

Production-grade SaaS covering the entire EU grant lifecycle

Structura delivered a production-grade SaaS platform that covers the entire EU grant lifecycle — from first idea to post-award compliance. Organisations can author applications faster with AI assistance, collaborate with stakeholders in real time, and manage funded projects without switching tools. The compliance module alone addresses a gap that no existing tool on the market fills properly.

This is our most complex project to date — 12 months, two runtime environments, 6+ AI pipelines, real-time collaboration, regulatory compliance, and a full project management suite. It demonstrates our ability to architect, build, and ship enterprise-grade AI SaaS products end to end — at any level of complexity.

Ready to build?

Want something like this?

Tell us about your project. We'll come back with a custom scope and proposal — no pressure.

Book a Free Discovery Call