Software engineers ready to ship LLM features.

Backend engineers, ML engineers, and senior developers building production systems who need to add generative AI capabilities responsibly.

Five-day syllabus

What you'll build.

DAY01

Foundations & prompt engineering

A practical mental model of how an LLM produces text — and the controls you actually have.

Tokenisation, context windows, attention — what changes when you change them
Decoding strategies: temperature, top-p, top-k, beam — when each matters
System / user / assistant message contracts; structured outputs
Few-shot, chain-of-thought, and prompt patterns that survive model updates

LabBuild a JSON-strict structured-output classifier with explicit validation, retry logic, and a prompt that gracefully degrades across three model providers.

DAY02

Tools, function calling, and agent loops

Move beyond chatbots: have the model take actions in your systems, safely.

Function calling across providers (OpenAI, Anthropic, local) — common ground
Tool schemas, argument validation, sandboxing, and audit trails
Agent loop architectures: ReAct, plan-and-execute, when not to agentise
Cost, latency, and failure-mode analysis for multi-turn agents

LabBuild a code-review agent that reads a Git diff, calls a linter and a test runner as tools, and produces a structured PR comment with citations to specific lines.

DAY03

Retrieval-augmented generation in practice

The most common production pattern — done with the rigour it deserves.

Embedding models compared; when fine-tuned embeddings actually help
Chunking strategies: fixed, semantic, hierarchical — and how to evaluate them
Hybrid search (BM25 + dense), reranking, and metadata filtering
Citation discipline: forcing the model to ground answers in retrieved passages

LabBuild a documentation-Q&A RAG over a real open-source codebase. Measure recall@k on a hand-built golden set; iterate the chunker until you beat the baseline.

DAY04

Evaluation, observability, and cost control

The discipline that separates a demo from a system you can put your name on.

Building golden datasets that survive prompt and model changes
LLM-as-judge: when it works, when it lies, and how to keep it honest
Tracing and observability: spans, prompts, costs, latencies, and replay
Caching layers, prompt compression, and choosing the right model per task

LabWire an evaluation harness around yesterday's RAG. Add tracing, cost-per-query metrics, and a CI check that fails the PR if quality drops below threshold.

DAY05

Safety, governance, and capstone

The realities you cannot ship without addressing.

Prompt injection & indirect injection; defences that actually work
PII detection and redaction in inputs and outputs
Abuse detection, rate limiting, and per-tenant isolation
Mapping your system to EU AI Act risk categories — practical exercise

CapstoneTake one of your real workplace problems and present an end-to-end design: data flow, prompts, retrieval, evaluation, monitoring, and a one-page risk register.

Lead instructor

Shady Ali

Co-Director · Sadiqoon Technologies

AI architect with deep production experience designing and operating sovereign LLM infrastructure for regulated industries. Teaches the patterns he ships.

What's included

Inclusions.

35 hours of live, instructor-led tuition
Full lab environment + cloud credits for the week
Printed bilingual workbook + digital code repository
Daily lunch & refreshments at the venue
Welcome dinner on Day 1; cohort dinner on Day 4
UK visa support letter on confirmed registration
30 days of post-course Q&A access to the instructor
Sadiqoon Institute Certificate of Completion

Reserve your seat.

Submit the form below and we will respond within two working days with a formal proforma invoice and the full registration pack.