EnableFlow AI

Independent portfolio project by Erica Nordlöf

EnableFlow AI is an internal AI enablement platform built as a portfolio lab. It demonstrates how AI can become a practical part of daily work across Operations, HR, Finance, Marketing, Sales, Product and Engineering.

The project mirrors the kind of responsibilities found in an AI Enablement Engineer role:

• Python and API integrations
• LLM integration
• RAG and internal knowledge search
• reusable prompts and workflows
• deterministic vs non-deterministic automation
• self-hosted AI concepts
• open-source models through Ollama
• commercial AI APIs through an OpenAI-compatible gateway
• vector database architecture with Qdrant
• Docker and self-hosted deployment
• documentation, governance and AI adoption

Important

This is an independent portfolio project. It demonstrates architecture, implementation and learning. It is not a claim that the author has already operated a company-wide AI platform in production.

Core principle

Not every task should use an LLM.

Known repeatable rule?        -> deterministic code
Language interpretation?      -> AI
Needs internal knowledge?     -> RAG
High-impact decision?         -> human review

Features

Internal knowledge assistant

Local company-like documents are indexed and retrieved before the model answers.

The demo uses a lightweight in-memory retrieval layer so it runs without a large embedding model. The architecture is prepared for Qdrant and a real embedding pipeline.

Workflow engine

Includes both:

• deterministic workflows
• AI-assisted workflows

Example: invoice variance approval remains deterministic because numeric thresholds should be reproducible and auditable.

Model gateway

Supports:

• demo mode with no external API key
• Ollama for local open-source models such as Llama or Qwen
• OpenAI-compatible commercial APIs

Reusable prompt library

Prompts are stored as reusable components with:

• ID
• department
• purpose
• variables
• template

Department use cases

Examples include:

• HR policy explanation
• Finance invoice exception review
• Marketing campaign brief drafting
• Sales account briefing
• Operations incident summaries
• Internal knowledge search

Run locally

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
uvicorn app.main:app --reload

Open:

http://localhost:8000

Demo mode works without API keys.

Docker

docker compose up --build

Optional local-model stack:

docker compose --profile local-ai up --build

Pull a model:

docker exec -it enableflow-ollama ollama pull qwen2.5:3b

Then use:

AI_PROVIDER=ollama
OLLAMA_MODEL=qwen2.5:3b

Architecture

Departments
    |
    v
FastAPI / Workflow API
    |
    +----------------------+----------------------+
    |                      |                      |
    v                      v                      v
Deterministic         Prompt / AI            RAG Service
Automation            Workflows                  |
                           |                      v
                           v                 Vector Layer
                     Model Gateway                |
                    /            \               v
             Commercial API    Ollama       Qdrant-ready
                                |
                           Llama / Qwen

Security ideas

• approved-provider boundaries
• no secrets in prompts or code
• human review for high-impact outputs
• audit metadata
• prompt injection awareness
• data classification before external AI use
• deterministic rules for auditable decisions

Portfolio description

EnableFlow AI is an independent AI enablement portfolio project combining Python, FastAPI, RAG, reusable AI workflows, model abstraction, self-hosted AI concepts, Docker and practical governance around AI adoption.

Tech

Python · FastAPI · REST API · RAG · Qdrant · Ollama · Llama/Qwen · Docker · JavaScript · Automation