Applied AI Automation for Developers
Software developers who want to build AI features and automations that work in production. You should be comfortable with one programming language, REST APIs and Git.
- Applied AI
- Intermediate
- 6 days, or 12 half-day sessions
- Instructor-led cohort · In-house for your team · Online, live
What you will be able to do
- Call large language model APIs from your own code, with structured output you can validate.
- Build a retrieval system that answers questions from your organisation's own documents and cites them.
- Give a model tools, such as looking up an order or creating a ticket, with limits on what it may do.
- Connect AI steps to real systems such as WhatsApp, email, an ERP like Odoo or a payment API.
- Measure answer quality with test sets and evaluations instead of guessing.
- Handle personal data, costs, failures and human review responsibly in production.
From a demo to something your business can run on
Getting a model to answer a question takes an afternoon. Getting it to answer correctly from your own data, act safely in your systems, stay within budget and fail gracefully takes engineering.
This course covers that engineering. It is the kind of work we do in our own AI automation practice: AI that connects to WhatsApp, an ERP and payment systems, with a person in the loop where it matters.
How it is taught
Each module is a short explanation followed by building. You work on one project throughout, a support assistant for a sample distributor that answers from its documents and looks up orders, and add retrieval, tools, evaluations and monitoring step by step.
Code examples are in Python and TypeScript. The patterns apply to any of the major model providers.
Syllabus
Working with LLM APIs
- Messages, system prompts, temperature and token limits
- Structured output and validating it with a schema
- Streaming, retries, timeouts and rate limits
Prompt design for software
- Prompts as code, with versions and reviews
- Few-shot examples and output formats
- Defending against prompt injection
Retrieval over your own documents
- Chunking documents and creating embeddings
- Vector search and keyword search, and combining them
- Answers with citations, and saying no when the answer is not in the documents
Tools and agents
- Function calling and tool design
- Keeping agents within limits, with approvals for risky actions
- When a plain workflow beats an agent
Integrating with business systems
- Webhooks and queues for WhatsApp and email automations
- Reading from and writing to an ERP through its API
- Workflow tools such as n8n, and when to write code instead
Evaluation and testing
- Building a test set from real questions
- Automated checks and model-graded evaluations
- Catching regressions when you change a prompt or a model
Running AI in production
- Logging, monitoring and cost control
- Personal data, consent and the Kenya Data Protection Act 2019
- Human review, fallbacks and handing over to a person
Assessment and certificate
- A practical project: you build a small AI automation that uses retrieval or tools against a sample business system, with an evaluation set, and explain its safeguards to an assessor.
- Deefrent Academy certificate, awarded on passing the course assessment.
- Ask for dates and fees. Tell us the course and the format you prefer, and we reply with the next dates and the fee.
Training a whole team?
This course can run privately for your organisation, at your premises or online, with examples drawn from your own systems.
Ready to join the next cohort?
Tell us how you would like to attend. We reply with the next dates, the fee and what you need before you start.