A field guide for Solution Architects & Product Owners

Architectural Traits of AI

Treat every model as a black box with four traits: its shape (what goes in, what comes out), its cost, its latency, and its behaviour with constraints. This guide organizes those traits into three pillars — and into the workflow you'll actually run every time a requirement lands on your desk.

The workflow: from requirement to production decision

Each step of the workflow is one pillar of the guide — this page is the map, the pillars are the tools.

What you do when a requirement lands

Requirement "detect defects in video" ① Match the shape I/O → family → model PILLAR 1 · SELECTION ② Size & cost it context · latency · VRAM PILLAR 2 · METRICS ③ Wire it in prompt · RAG · fine-tune PILLAR 3 · INTEGRATION Evaluation gate precision · recall · cost/request vs the expensive error pass Ship it monitored · budgeted · fallbacks fails the gate? change the family or the pattern — not just the prompt
Note what's absent: no training, no math, no GPUs of your own. The architect's job is selection, sizing and wiring — the pillars exist to make each step a 10-minute decision instead of a research project.
Your ~10% of math is operational, not algorithmic — token budgets, VRAM arithmetic, latency targets, reading an evaluation report (all in pillar 2). Everything under the hood — losses, gradients, distributions — belongs to the ML Engineer's track. Know what a GAN does and when to use it, not how to compute its Nash equilibrium.

The three pillars

Each pillar has its own detail page with decision diagrams, traits and cheat sheets.

The questions you'll actually be asked

Stakeholder question → your move → where the answer lives.

When they ask…The architect's moveWhere
"Can AI solve this?" Translate the requirement into an I/O shape, then match it to a family Model selection — decision diagram
"Which model should we shortlist?" Pick the variant inside the family — with named models per type Family deep dives
"What will it cost to run?" VRAM & token arithmetic: params × bytes, tokens × price, headroom Operational metrics — VRAM & context
"Why is the pilot so slow?" Separate latency from throughput; check TTFT and workload mixing Operational metrics — latency
"Can it use our internal data — securely?" RAG with access control enforced at the retrieval step Integration — RAG architecture
"Should we fine-tune?" Walk the escalation ladder: prompt → RAG → fine-tune, only on proven failure Integration — the ladder
"DS reports 97% accuracy — can we ship?" Ask which error is expensive, then check that metric, not the average Operational metrics — evaluation

Quick orientation: which family for which job

The 10-second version of pillar 1 — full traits and examples on the detail page.

Business requirementReach forTypical models
Understand / generate text, answer questionsTransformer LLM (+ RAG for your data)GPT-4, Claude, Llama
Detect objects or defects in images / videoCNN-based vision modelYOLO, ResNet
Generate images from textDiffusion modelStable Diffusion, DALL·E
Synthetic data, upscaling, face generationGANStyleGAN, ESRGAN
Predict from business tables (churn, fraud, credit)Classic ML — gradient boostingXGBoost, LightGBM
Semantic search, similarity, clusteringEmbedding model + vector DBtext-embedding models, CLIP

Go deeper per family — types with named models

← All field guidesArchitect · Engineer · Scientist