IN PROGRESS RAG · Retrieval

PflegeLotse

GDPR-compliant RAG assistant for German long-term-care law (SGB XI) — source-grounded answers and an eval harness.

Live-Demo (soon) Code (soon)

Problem & context

Care law is complex — and answers must be verifiable

Care providers and families lose time searching the SGB XI. A generic chatbot hallucinates — unacceptable for legal questions. What is needed is an assistant that answers only from the statute, cites every claim, and abstains honestly when unsure.

Solution

Retrieval-augmented generation with strict grounding

Hybrid retrieval via pgvector, source-grounded answers, anti-hallucination and abstention.

Screenshot / demo GIF coming

Architecture

Clean Architecture, four layers

domain

Entities & rules — framework-free

application

Use cases: ingest, retrieve, answer

infrastructure

pgvector, E5 embeddings, Mistral/Ollama

api

FastAPI + Jinja2/HTMX

Process history

From plan to deploy — six phases

  1. 01

    Setup & architecture

    IN PROGRESS

    Clean-architecture skeleton, Docker, CI. ADR-0001: Python/HTMX over Next.js.

  2. 02

    Data & ingestion

    PLANNED

    Load SGB XI (public domain), chunk, E5 embeddings → pgvector.

  3. 03

    Retrieval & grounding

    PLANNED

    Hybrid retrieval, grounded answers, abstention.

  4. 04

    Eval harness

    PLANNED

    Measure recall@k, faithfulness, abstention.

  5. 05

    UI & HITL

    PLANNED

    Jinja2/HTMX, source display, disclaimer (RDG).

  6. 06

    Deploy & docs

    PLANNED

    Docker deploy, README with metrics, ADRs, datenschutz.md.

Results

Made measurable

—
Recall@k
—
Faithfulness
—
Abstention rate

Will be filled with real numbers after the eval phase — and then feeds into the CV.

Stack & compliance

Python 3.12FastAPIpgvectorE5-EmbeddingsMistral / OllamaHTMXDocker

GDPR & EU AI Act: no personal data, EU or local LLM, citations instead of free generation. Disclaimer: no legal advice (RDG).

PflegeLotse live demo

← All projects