Thank you for sending your enquiry! One of our team members will contact you shortly.
Thank you for sending your booking! One of our team members will contact you shortly.
Duration 35 hours
Course Outline
1. LLM Architecture and Core Techniques
- Analysis of distinctions between Decoder-Only (GPT-style) and Encoder-Decoder (BERT-style) architectures.
- Examination of Multi-Head Self-Attention mechanisms, positional encoding protocols, and dynamic tokenization processes.
- Application of advanced sampling strategies, including temperature adjustment, top-p filtering, beam search, logit bias, and sequential penalties.
- Comparative assessment of leading models such as GPT-4o, Claude 3 Opus, Gemini 1.5 Flash, Mistral 8×22B, LLaMA 3 70B, and quantized edge variants.
2. Enterprise Prompt Engineering
- Implementation of prompt layering strategies, including system prompts, context prompts, user prompts, and post-prompt processing workflows.
- Utilization of Chain-of-Thought, ReACT, and auto-CoT techniques incorporating dynamic variables.
- Design of structured prompts utilizing JSON schemas, Markdown templates, and YAML function-calling specifications.
- Mitigation of prompt injection risks through sanitization, length constraints, and default fallback protocols.
3. AI Tooling for Developers
- Overview and comparative application of GitHub Copilot, Gemini Code Assist, Claude SDKs, Cursor, and Cody for government software development.
- Best practices for integrating AI tools within IntelliJ (Scala) and VSCode (JS/Python) environments.
- Cross-language benchmarking for coding, test generation, and refactoring tasks to ensure efficiency and accuracy.
- Customization of prompts per tool using aliases, contextual windows, and snippet reuse for consistent output.
4. API Integration and Orchestration
- Implementation of OpenAI Function Calling, Gemini API Schemas, and Claude SDK integrations for end-to-end workflows.
- Management of rate limiting, error handling, retry logic, and billing metering for reliable service operation.
- Development of language-specific wrappers:
- Scala: Akka HTTP
- Python: FastAPI
- Node.js/TypeScript: Express
- Integration of LangChain components, including Memory, Chains, Agents, Tools, multi-turn conversation management, and fallback chaining.
5. Retrieval-Augmented Generation (RAG)
- Parsing of technical documents in formats such as Markdown, PDF, Swagger, and CSV using LangChain and LlamaIndex.
- Application of semantic segmentation and intelligent deduplication techniques to enhance data integrity.
- Utilization of embedding models including MiniLM, Instructor XL, OpenAI embeddings, and Mistral local embedding.
- Management of vector stores such as Weaviate, Qdrant, ChromaDB, and Pinecone, with focus on ranking and nearest-neighbor tuning.
- Implementation of low-confidence fallback mechanisms to alternate LLMs or retrievers to ensure continuous service availability.
6. Security, Privacy, and Deployment
- Enforcement of PII masking, prompt contamination control, context sanitization, and token encryption for data protection.
- Establishment of prompt and output tracing protocols, including audit trails and unique identifiers for each LLM call.
- Configuration of self-hosted LLM servers (Ollama + Mistral), GPU optimization, and 4-bit/8-bit quantization for resource efficiency.
- Deployment via Kubernetes using Helm charts, autoscaling, and warm start optimization to support high-availability requirements.
Hands-On Labs
- Prompt-Based JavaScript Refactoring
- Multi-step prompting workflow: detection of code smells, proposal of refactors, generation of unit tests, and creation of inline documentation.
- Scala Test Generation
- Creation of property-based tests using Copilot versus Claude, with measurement of coverage and edge-case generation.
- AI Microservice Wrapper
- Construction of a REST endpoint that accepts prompts, forwards them to LLMs via function-calling, logs results, and manages fallback logic.
- Full RAG Pipeline
- Processing of simulated documents through indexing, embedding, retrieval, and search interfaces with ranking metrics.
- Multi-Model Deployment
- Containerized setup with Claude as the primary model and Ollama as a quantized fallback, monitored via Grafana with defined alert thresholds.
Deliverables
- Shared Git repository containing code samples, wrappers, and prompt tests for government application.
- Benchmark report detailing latency, token cost, and coverage metrics for performance evaluation.
- Preconfigured Grafana dashboard for monitoring LLM interactions and system health.
- Comprehensive technical PDF documentation and a versioned prompt library for long-term use.
Troubleshooting
Summary and Next Steps
Requirements
- Proficiency in at least one programming language (Scala, Python, or JavaScript).
- Familiarity with Git, REST API design, and CI/CD workflows.
- Basic understanding of Docker and Kubernetes concepts.
- Commitment to applying AI/LLM technologies to enterprise software engineering for government purposes.
Target Audience
- Software Engineers and AI Developers
- Technical Architects and Solution Designers
- DevOps Engineers implementing AI pipelines
- R&D teams exploring AI-assisted development
Testimonials (2)
The course was very useful, and the trainer was clear, well-prepared, and engaging. I liked the fact that it was very practical with labs and real use cases. Overall, it was a valuable training experience.
Mattia Dettori - MFM INVESTMENT Ltd Italian branch
Course - LLM Engineering Bootcamp
1. The section on the constraints, tests, and guardrails was cool. 2. Francesco knows the topic very well, and he’s a nice teacher. 3. Overall, I really enjoyed the course, and I learnt a lot.
