GenAI App Development: LLM APIs, RAG and Vector Search for Developers
Calling a language model from code is easy. Building an application that gives reliable, secure and affordable answers from your own data is harder. This two-day course gives developers the patterns for doing it well, in a provider-neutral way.
Participants build a small retrieval-augmented application step by step, covering prompts, embeddings, vector search, evaluation and security. Exercises use Python. The course complements the OWASP LLM Top 10 course, which goes deeper on security, and the MLOps course, which covers classical models.
HRD Corp SBL-Khas Claimable
Course at a Glance
| Duration | 2 days |
|---|---|
| Delivery | Physical or online. Tell us your preference when you enquire. |
Programme Agenda
Day 1, 9:00 AM - 9:15 AM
Welcome and Programme Overview
Introduction to the session, objectives, and housekeeping.
Day 1, 9:15 AM - 10:15 AM
How LLM Applications Are Put Together
Models, APIs, prompts, context windows, tokens and cost. The parts of a typical application, what the model is good and bad at and how to choose between hosted and self-hosted models.
Day 1, 10:15 AM - 10:30 AM
Break
Day 1, 10:30 AM - 12:30 PM
Working With an LLM API
Sending requests, system and user messages, parameters, structured output, errors, rate limits and streaming. Keeping keys and personal data safe.
Day 1, 12:30 PM - 1:30 PM
Lunch
Day 1, 1:30 PM - 3:15 PM
Prompt Design and Output Control
Writing instructions that hold up, few-shot examples, asking for a fixed format, validating outputs in code and handling refusals and drift.
Day 1, 3:15 PM - 3:30 PM
Break
Day 1, 3:30 PM - 4:45 PM
Embeddings and Vector Databases
What an embedding is, how similarity search works, chunking documents, metadata and choosing and querying a vector store.
Day 1, 4:45 PM - 5:00 PM
Day 1 Close
Recap and what to review before day two.
Day 2, 9:00 AM - 9:15 AM
Day 1 Review
Recap of day one and the questions it left open.
Day 2, 9:15 AM - 10:15 AM
Retrieval-Augmented Generation
Retrieving relevant passages, building the prompt from them, citing sources, handling missing answers and improving retrieval through chunking, filters and re-ranking.
Day 2, 10:15 AM - 10:30 AM
Break
Day 2, 10:30 AM - 12:30 PM
Evaluation and Quality
Building a small test set, measuring answer accuracy, grounding and cost, spotting regressions after changes and the limits of automated judges.
Day 2, 12:30 PM - 1:30 PM
Lunch
Day 2, 1:30 PM - 3:15 PM
Security, Privacy and Safety
Prompt injection, data leakage, access control on retrieved documents, logging without storing personal data and the PDPA. Links to the OWASP LLM Top 10.
Day 2, 3:15 PM - 3:30 PM
Break
Day 2, 3:30 PM - 4:45 PM
Deployment and Operations
Caching, latency, cost control, monitoring, versioning prompts and models, and a handover checklist. Participants present their working application.
Day 2, 4:45 PM - 5:00 PM
Wrap-Up and Q&A
Key takeaways, next steps, and close.
Key Outcomes
- Describe the parts of an LLM application and where cost and risk arise
- Call an LLM API with structured, validated output
- Prepare documents and build embedding-based search
- Build a retrieval-augmented answer flow with citations
- Evaluate quality with a small test set
- Apply basic security, privacy and operational controls
Training Mode Physical / Online / Hybrid / e-learning
HRD Corp SBL-Khas Claimable
Level Intermediate, for developers with working knowledge of Python. Prior machine learning experience is not required.
Duration 2 Days (16 Hours) | 9:00 AM to 5:00 PM daily
Venue In-house at the client's premises, or delivered via the client's preferred platform (Microsoft Teams, Zoom, or equivalent)
Assessment A working retrieval-augmented application built during the course, plus a short knowledge check
Certificate Certificate of Completion issued to all participants upon full attendance