Now enrolling at Chelsea World School · 2026–27 · Seats limited per bandEnquire now →
Code Camp by SCCCode Camp by SCCNow at Chelsea
AI PathwayAt ChelseaHow We TeachStudent WorkFor Schools
Enrol →
← The whole climb
Classes 11-12 · Year one • Pumori SummitEngineering-grade

Engineer the system end to end.

The problem and operating context are supplied. The architecture is not. Pumori asks the learner to independently engineer the technical solution: algorithms, data structures, services, databases, pipelines, models, LLM components, bounded agents, tests, deployment behavior and reliability.

“I can engineer substantial systems with evidence.”

Who owns whatThe problem and operating context are supplied; the learner independently designs the architecture and owns the technical implementation, evaluation and reliability.

  1. Classes 4–5 · Year oneMera Basecamp
  2. Classes 4–5 · Year twoMera Peak
  3. Classes 6–7 · Year oneLobuche Basecamp
  4. Classes 6–7 · Year twoLobuche Summit
  5. Classes 8–9 · Year oneAma Dablam Basecamp
  6. Classes 8–9 · Year twoAma Dablam Summit
  7. Classes 11–12 · Year onePumori Summit
  8. Classes 11–12 · Year twoCho Oyu Summit

“I can formulate and build mature components and systems.”

→

“I can independently architect and engineer a substantial intelligent system for a serious defined problem.”

What changes this year

  • Complexity reasoning becomes explicit: O(1), O(log n), O(n), O(n log n), O(n²) at an applied level.
  • Data structures expand to stacks, queues, heaps, trees, graphs, matrices and richer records/objects.
  • Students enter unfamiliar repositories and trace behavior across multiple modules and dependencies.
  • They design the software architecture for the defined problem.
  • They consume and create APIs and use relational databases/SQL.
  • Multi-source data pipelines, provenance, joins and leakage become first-class concerns.
  • ML includes model comparison/optimization and practical unsupervised learning.
  • LLM systems include embeddings, retrieval/RAG, structured output and tool calling.
  • Bounded agents must have tools, permissions, validation, stopping criteria and human approval where needed.
  • Reliability means designing how the system fails: retries, timeouts, fallbacks, logs and escalation.

AI engineering, not chatbot assembly

Class 11 begins with reactivation, not amnesia.

After the Class 10 gap, the system retrieves and diagnoses important prior capabilities, repairs gaps and restores fluency before accelerating into advanced work.

Try the thinking

A minute of Pumori Summit thinking.

An illustrative interaction, not a lesson from the curriculum — but the reasoning is the reasoning.

Every layer says 'success.' The answer is still wrong.

Interactive example

A serious intelligent system fails across boundaries, not only inside one function.

  1. →
  2. →
  3. →
  4. →
  5. →

Where would you investigate first?

Inspect at least 2 components first. The layer that reports a failure is not automatically the layer that caused it.

What they produce

  • Architecture diagrams
  • Service/API contracts
  • Relational schemas and SQL
  • Multi-source data pipeline documentation
  • Model comparison/evaluation reports
  • RAG/tool-calling system components
  • Bounded agent with explicit permissions and stopping rules
  • Automated tests and reliability evidence
  • Deployment + technical README

What you should be able to see

  • ✓An architecture the learner designed
  • ✓A data-structure/algorithm choice with complexity reasoning
  • ✓SQL/schema and data-pipeline evidence
  • ✓A model baseline, comparison and error analysis
  • ✓An LLM/retrieval or tool-calling component evaluated separately
  • ✓A bounded agent with explicit allowed actions and human-control points
  • ✓A cross-component bug the learner diagnosed
  • ✓A reliability/security review

Support matters. A heavily guided success is not reported as independent mastery.

Twelve Projects. One climb.

A Band is not a pile of unrelated activities. Each standard Band contains 12 Projects arranged so that capability is established, combined, transferred and increasingly owned.

  1. P1–P4Establish / Acquire
  2. P5–P8Combine / Extend
  3. P9–P12Transfer / Own
Illustrative synthesis example

A bounded operations system combining data, storage, ML/LLM components and approved tool actions

The student receives a serious bounded brief, stakeholders, requirements, constraints and initial data—but no architecture. The system must include meaningful algorithm/data choices, persistent storage, a pipeline, model/AI components, evaluation and reliability behavior.

Design archetype: Engineer an Intelligent Operations System

Illustrative synthesis archetype from current Band design; not a guaranteed final Project title.

Where Pumori Summit lands on each capability

These are Band end-states, not lesson titles. Follow any of them across all eight years in the Capability Telescope.

AAlgorithms & problem solving
Choose/adapt BFS, DFS, shortest-path, recursion, backtracking, greedy, introductory dynamic programming or heuristic search and reason explicitly about complexity.
BProgramming & representation
Enter repositories with multiple modules, dependencies, APIs and databases; locate entry points and safely modify behavior.
CDebugging
Trace a failure across UI → API → database → retrieval/model → parser and use logs, request inspection and controlled experiments.
DTesting & reliability
Use unit/integration/system/regression thinking, automated execution, fallbacks, retries, timeouts and AI evaluation datasets.
EData & evidence
Build multi-source pipelines, reason about schemas, joins, provenance, leakage, feature creation and model metrics.
FAI & machine learning
Compare/optimize supervised models, use unsupervised learning, build LLM systems with embeddings/retrieval/RAG/tool calling and engineer bounded agents with reliability controls.
GSystems & architecture
Independently design UI, services, databases, external APIs, ML/LLM/retrieval components, background work and evaluation/logging for a defined problem.
HTechnical judgment
Compare deterministic functions, fixed workflows and agents; reason about quality gain vs complexity, latency, cost, security and failure.
ISecurity, privacy & human-centered computing
Use roles, least privilege, secrets, input validation and AI-specific safeguards such as tool permissions and prompt-injection awareness.
JExplanation, transfer & technical defense
Create architecture/data-flow/schema/API/evaluation documentation and defend the end-to-end system.
Open the Capability Telescope→
What becomes possible next

What Cho Oyu adds

Pumori owns almost the entire technical problem. Cho Oyu adds the problem itself: what should exist, for whom, under which constraints, with which automation and with what definition of success.

Climb to Cho Oyu Summit
← What prepared them: Class 10 ContinuitySee the whole climb

Program

  • AI Pathway
  • How We Teach
  • Student Work
  • At Chelsea World School
  • For Schools

Company

  • About
  • FAQ
  • Contact
  • Privacy

Social

  • Samman Creative
  • Facebook

Get Started

© 2025 Samman Creative and Consulting — All rights reserved.

Built and run by SCC

Built with love from Kathmandu.

For DesignersTerms of usePrivacy policy
💬Enquire · WhatsApp