Full-Stack Engineer: Self-Hosted Workflow + RAG App

We're hiring an expert full-stack engineer to build an MVP of a privacy-first, self-hosted AI web app.

The product is a guided, structured experience (not a generic chatbot): users go through a step-by-step workflow and the system produces live-updating deliverables during the session (tables/notes/frameworks) plus a downloadable PDF report.

Language:

UI + outputs are primarily German.

What you'll build (MVP)

Core UX

Two-panel web app:

Left:

chat-guided workflow ("orchestrator")

Right:

live "artifact" panel that updates in real time (structured content, tables, summaries), plus report preview/export

Core capabilities

Orchestration / workflow layer (self-hosted)

Routes user intents to specialized sub-flows (ingestion, retrieval, artifact creation, reporting)

Document ingestion + retrieval (RAG)

Upload PDFs and spreadsheets

Detect text-based vs scanned docs, OCR only when needed

Chunking + embeddings + retrieval against a vector store

Hard multi-tenancy

Strict isolation per tenant so cross-tenant retrieval is impossible by design (not just "by convention")

Real-time updates

Streaming chat responses and live artifact updates (SSE or WebSockets)

Templated PDF report generation

Generate a branded PDF using templates via a dedicated service

Admin-configurable "skills"

Upload instruction/config files (JSON/YAML) that adjust behaviour/workflows without redeploying core code

Telemetry & cost tracking

Log model calls + usage (tokens/costs) and workflow events per tenant

Testing / review dashboard

Simple UI to test flows, inspect retrieved context chunks, see configuration used, and record thumbs up/down feedback

Hosting & security requirements (non-negotiable)

Must run self-hosted in a controlled environment (containerized deployment is expected).

Databases/services must not be publicly exposed.

Proper secrets handling (no hardcoded keys).

Clear service boundaries and basic hardening best practices.

Stack (flexible)

You can propose your preferred implementation choices as long as the above requirements are met

Typical building blocks include:

A modern frontend framework (React/Vue or similar)

Backend in or Python (or equivalent)

Relational DB + vector DB

A self-hosted orchestration/workflow engine

LLM inference via a major API provider (inference only)

Deliverables

Working MVP deployed via containers with clear environment configuration

Chat + artifact UI with file upload, streaming, and PDF export

Tenant isolation implemented and verified with leakage tests

Usage logging + feedback logging

Admin skill/config loader

Testing dashboard

Documentation:

Architecture diagram

Deployment guide ("fresh server to running")

Runbook (ops + troubleshooting)

How to add/update a skill via config files

Engagement style

Prefer milestone-based delivery with clear acceptance criteria

Short weekly check-in + async updates

Clean repo, readable code, and handover-ready docs

Contract duration of 3 to 6 months.

Mandatory skills:

React, Docker, Chatbot Development, API, API Integration, JavaScript,