IN PROGRESS Agents · Tool-calling

PostLotse

AI agent for SME email triage — categorises, prioritises and drafts replies, with a human in the loop.

Live-Demo (soon) Code (soon)

Problem & context

Email overload costs SMEs hours every week

Requests, invoices, appointments and spam land in one inbox. An agent that classifies, prioritises and drafts routine replies — escalating tricky cases — gives that time back.

Solution

An agent with tools — not a chatbot

Tool-calling (set label, draft reply, create appointment) in an agent loop with an approval step.

Screenshot / demo GIF coming

Architecture

Clean Architecture, four layers

domain

Triage classes & escalation rules

application

Agent loop, tool orchestration, HITL

infrastructure

IMAP, Mistral, tool registry

api

FastAPI + HTMX dashboard

Process history

From plan to deploy — six phases

  1. 01

    Setup & architecture

    IN PROGRESS

    Clean-arch skeleton, Docker, CI. ADR-0001: tool-calling agent over rigid rules.

  2. 02

    Classification

    PLANNED

    Categories, few-shot prompts, measure baseline.

  3. 03

    Tool-calling

    PLANNED

    Tools exposed, agent loop with Mistral.

  4. 04

    Human-in-the-loop

    PLANNED

    Escalation + approval before any action.

  5. 05

    Eval harness

    PLANNED

    Routing F1, tool-call success, escalation rate.

  6. 06

    Deploy & docs

    PLANNED

    Docker deploy, README, ADRs, datenschutz.md.

Results

Made measurable

—
Routing F1
—
Tool-call success
—
Escalation rate

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

Stack & compliance

Python 3.12FastAPIHTMXIMAPMistralTool-CallingDocker

GDPR: emails are personal data → EU/local LLM, data minimisation. Human-in-the-loop prevents uncontrolled actions.

PostLotse live demo

← All projects