EB-1
The discipline of providing the right information, tools, and memory, in the right format, before the model reasons. If the model is the engine, context is the fuel mixture — and getting it wrong is the single most common cause of unreliable agents.
Theme: Context →19 topics — scope and references for each.
Context engineering is the successor concept to prompt engineering. Where prompt engineering asks *how do I phrase the request*, context engineering asks *wh…
Chain-of-Thought (CoT) is the foundation every other reasoning pattern builds on: eliciting step-by-step reasoning either through explicit worked examples or…
ReAct (Reason + Act) is the pattern that made tool-using agents practical and remains the default agent loop in most frameworks. It interleaves reasoning tra…
A practical framework for managing what enters the context window. **Write** stores information outside the window (scratchpads, runtime state, external note…
Retrieval-Augmented Generation is the workhorse of the "Select" strategy. Students learn the pipeline end to end: chunking source material, generating embedd…
A reliable prompt is structured, not improvised. Students learn to organise prompts into clearly delimited sections — background, instructions, tool guidance…
Two planning-oriented alternatives to ReAct's step-by-step loop. **Plan-and-Execute** generates a full plan upfront and then executes each step; it is effici…
A decision framework rather than a single technique. The same agent can be built four different ways, each with a different failure mode and cost profile. St…
Reflexion lets an agent improve through self-feedback without any retraining. After a failed or completed attempt, the agent writes a natural-language critiq…
Where CoT commits to a single linear chain, Tree of Thoughts explores several reasoning branches at once and evaluates them, dramatically outperforming linea…
Long-running agents accumulate context that degrades their performance — earlier information receives less attention, and irrelevant material pollutes reason…
A direct application of the "Isolate" strategy at system scale. Splitting work across several agents, each with its own focused context window, can outperfor…
Beyond naïve retrieval: hybrid search combining semantic and keyword matching, re-ranking with cross-encoders, metadata filtering, and handling cases where t…
The conceptual spine of the Advanced track. Human cognition falters under overload because working memory is sharply limited — the central claim of Cognitive…
CLT distinguishes intrinsic load (the irreducible difficulty of the task), extraneous load (difficulty added by poor presentation), and germane load (effort…
A formal account of *why* long contexts degrade. Research operationalises two mechanisms: **context saturation**, where extraneous task-irrelevant informatio…
The human/machine parallel has a security consequence. Researchers have shown that, just as humans make errors under overload, LLMs can be deliberately pushe…
Where the parallel breaks. Metacognition — thinking about one's own thinking — is a key point of divergence: humans monitor and regulate their own reasoning…
Metacognition can be partially induced in the model itself. Metacognitive prompting structures a response into stages that emulate human self-monitoring — un…
Topics are combined ad hoc for corporate programs and workshops — there is no fixed public duration. Tell us your goals and we design the track.
Discuss a custom program