One focused AI-assisted workflow for a defined repetitive task, with limited integrations and explicit human review.

Defined package • One trigger, up to five processing steps and up to two supported app or API integrations. • One existing AI model for classification, extraction or draft generation. • Input validation, basic failure handling, sample-based testing and human approval before external communication where applicable. • One agreed environment setup, documentation and handover.

Working example Our internal lead demonstration uses Python, Ollama, Qwen 2.5 and SQLite. It captures enquiries, proposes qualification, records a local CRM entry and gates follow-up drafts behind human approval. Fictional data only. Approved drafts enter a local outbox; no real emails were sent and no commercial CRM is connected. This is a functional prototype, not a client commission or production-ready platform.

Delivery: approximately 1–2 weeks after requirements, access and dependencies are confirmed, with two consolidated revision rounds. API access, hosting, subscriptions and model usage are separate. Additional agents, complex branching, model training, production hardening and ongoing monitoring require a custom quotation. AI output remains provisional and subject to review.