CP3405DT3 · System Intelligence Platform
Prof. Dr. Tan · TR2 '26
SPRINT W3 · SYSTEM ARCHITECTURE PAGE · Manual → Structured → Automated → Intelligent
SYSTEM THINKING

CP3405 System Architecture

Scrum does not change. The complexity of the system changes. Your team is not completing tasks — you are designing an intelligence pipeline.

Core Principle

Manual Work Is a System Design Signal

Every time your team does something manually — collecting data, running prompts, comparing outputs, scoring evidence — you are discovering a future automation requirement. This is not inefficiency. This is system specification through lived experience.

Strong teams don't just complete the sprint. They ask: how would we build this so a machine could do it at scale?

Diagram 1 — Intelligence Pipeline + Scrum Ceremony Map

Data Sources → Agents → LLM Synthesis → Human Score → Lock → Delta → Prescribe → Repeat

This is the complete Scrum-aligned sprint cycle. The left half is Sprint Execution (Planning, Backlog Refinement, Daily Scrum). The right half — from Locked Decision onward — is the Sprint Review and Retrospective codified as a system. The teal calibration loop at the bottom is Scrum's Inspect and Adapt principle made visible: every Rx prescription is a retrospective action item that enters next sprint's backlog. You are not building a trading system. You are building a disciplined system improvement engine.

SPRINT EXECUTION SPRINT REVIEW + RETROSPECTIVE 📡 DATA SOURCES Finviz · Yahoo TradingEcon · Charts Sprint Planning inputs 🔬 AGENTS Almanac R3 · Macro R4 Technical R5 · Data R6 Backlog refinement 🤖 LLM SYNTHESIS GPT · Claude Gemini · DeepSeek Planning Poker 🧠 HUMAN SCORE R7 Override Wild Card Insight Product Owner 🔒 LOCKED GitHub vWN Tag Discord · Deck Definition of Done Δ DELTA ENGINE Predicted vs Actual Direction · Size Sprint Review Rx PRESCRIBE Agent re-weights New signals Retro → Backlog R10 CALIBRATION — Inspect and Adapt Rx prescriptions feed next sprint backlog SCRUM CEREMONY MAPPING — This pipeline IS your Agile practice, not separate from it SPRINT PLANNING What data do we need? What is the sprint goal? BACKLOG REFINEMENT Each agent = specialist. No groupthink before LLMs. PLANNING POKER 4 LLMs = 4 estimates. Divergence = hidden risk. PRODUCT OWNER R7 protects value, not just speed. DEFINITION OF DONE Tag seals prediction. No retroactive edits. SPRINT REVIEW Inspect the increment. Delta = learning, not blame. SPRINT RETRO Rx = action items. Enter next sprint backlog.
Diagram 2 — Layered Architecture

Sources → Ingestion → Storage → Processing → Output

A software architecture view of how Prism — your future tool — would be structured. Each layer your team touches manually this sprint will eventually become code.

OUTPUT LAYER
SPX Prediction
NDX Prediction
IWM Prediction
Monday Presentation
GitHub Release
PROCESSING
Human Score Engine (R7)
LLM Comparison Engine
Confidence Scoring
Invalidation Logic
INTELLIGENCE
ChatGPT API
Claude API
Gemini API
DeepSeek API
FinBERT (Sentiment)
AGENT LAYER
Almanac Agent R3
Macro Agent R4
Technical Agent R5
Data / Actuals R6
STORAGE
GitHub Repository
Evidence Folder
FigJam Thinking Wall
Discord Thread
INGESTION
Data Collector (Manual → Scripted)
Screenshot / URL Evidence
Market Calendar Parser
SOURCE LAYER
Finviz Sector Map
Yahoo Finance 5D
TradingEconomics
Chart Evidence
SPX · NDX · IWM Live
Diagram 3 — System Maturity

Level 1 → Level 2 → Level 3 → Level 4

Honest assessment: most teams are currently at Level 1–2. The goal of this course is to get you thinking and building toward Level 3–4.

🛠️ LEVEL 1 MANUAL Everything by hand. High effort. Low scale. Prone to errors. ← MOST TEAMS NOW 📋 LEVEL 2 STRUCTURED Templates & roles. GitHub evidence trail. Repeatable process. ← TARGET FOR THIS SPRINT ⚙️ LEVEL 3 AUTOMATED Scripts, APIs, jobs. Human reviews output, not inputs. ← COURSE ASPIRATION 🧠 LEVEL 4 INTELLIGENT AI synthesis, anomaly detection. Self-improving predictions. PRISM. ← WHERE PRISM LIVES
System Thinking Principles

What Strong Teams Understand

Manual = System Signal
Every repetitive manual step is a future automation requirement. Do not accept friction as the natural state.
You Are Building Systems
This course is not about completing assignments. It is about designing, testing, and improving a market intelligence system.
Automation Frees Judgement
When data collection is automated, human attention moves to where it matters: interpretation, scepticism, and the Wild Card.
Scale Test Everything
Ask: "What breaks when we scale to 10 teams?" If the answer is "everything", your process is not a system yet.
Scrum Exposes the System
Sprints, standups, retrospectives, and release tags are not ceremony. They are the feedback loop that makes the system self-correcting.
Human Score = Irreplaceable
Four LLMs cannot replace the R7 Human Score. Pattern recognition, news intuition, and contrarian thinking remain human advantages.
Architecture Deep Dive →