Skip to content

Lesson 365·M29 Enterprise AI Solutions Architecture

Model Selection at Scale

Open

Learning Objectives

By the end of this lesson you should be able to:

  • Explain the tiers: the small, the mid, the frontier (L365)
  • Explain the routing: the tier per request (L155)
  • Explain the fine-tuning: the custom model (L365)
  • Explain the economics: the cost and the quality (L334)
  • Explain the AI shape: the tiered strategy (L365)

One-Line Definition

The model selection at scale is the routing, the fine-tuning, and the tiered model strategy for the enterprise (L365) — the tiers (the small L157: the cheap and the fast L151; the mid: the balance; the frontier L148: the best and the costly L150, L365), the routing (the tier per request: the simple L365 to the small L157, the hard L365 to the frontier L148, L155), and the fine-tuning (the custom L365: the domain L365, the format L365, L365) — the scale's (L358) model economics (L334).

The one-sentence interview answer: "The model strategy at the scale is the tiers, the routing, and the fine-tuning (L365). The tiers (L365): the small (L157) — the cheap (L150) and the fast (L151): the classification (L353), the extraction (L143); the mid (L365) — the balance: the summarization (L166); and the frontier (L148) — the best (L341) and the costly (L150): the complex reasoning (L203). The routing (L155): the tier per request (L365) — the simple (L365) to the small (L157), the hard (L365) to the frontier (L148) — the classifier (L353) routing (L155) — the cost (L334) and the latency (L333) balanced (L365). The fine-tuning (L365): the custom (L365) — the domain (L365): the legal (L371), the medical (L371) — the format (L365): the structured (L143) — the fine-tuned (L365) model (L148) for the high-volume (L358) specific (L365) task (L365). The economics (L334): the tokens (L332) per tier (L365) — the L334 attribution (L334) — the tier's (L365) cost vs the quality (L341). The AI shape (L173): the enterprise (L380) — the models (L148) tiered (L365), routed (L155), and fine-tuned (L365) — the scale's (L358) strategy (L365)."

Mental Model

Think of the model strategy as the law firm's partner tiers. The firm (the enterprise, L380) has the tiers (L365): the associates (the small, L157) — the cheap (L150) and the fast (L151): the filings (the classification, L353); the partners (the mid, L365) — the balance: the briefs (the summarization, L166); and the senior partners (the frontier, L148) — the best (L341) and the costly (L150): the trials (the reasoning, L203). The intake (the routing, L155): the case's (the request's, L365) complexity (L365) → the tier (L155). And the specialists (the fine-tuning, L365): the domain experts (L365) — the tax (L371), the patent (L371) — hired (the fine-tuned, L365) for the firm's (L358) high-volume (L358) niche (L365). The firm works because the tiers match the cases, the intake routes, and the specialists handle the niches (L365).

   the firm (the strategy, L365)
   ┌────────────────────────────────────────────────────────┐
   │ the tiers (L365) — the associates (L157), the partners │
   │ (L365), the seniors (L148)                             │
   │ the intake (the routing, L155) — the complexity → the  │
   │ tier (L365)                                            │
   │ the specialists (the fine-tuning, L365) — the niches   │
   └────────────────────────────────────────────────────────┘

The mental model is the firm: the tiers, the intake, and the specialists (L365).

Visual Flow — One Request Through the Tiers

   the request (L365)
        │
        ▼
   ┌────────────────────── THE ROUTER (L155) ───────────────────────────┐
   │  the classifier (L353): the complexity (L365)                     │
   │  the simple (L365) → the small (L157)                             │
   │  the mid (L365) → the mid (L365)                                  │
   │  the hard (L365) → the frontier (L148)                            │
   └──────────────────────────┬───────────────────────────────────────┘
                              ▼
   ┌────────────────────── THE TIER (L365) ─────────────────────────────┐
   │  the small (L157): the cheap, the fast (L151)                     │
   │  the mid (L365): the balance (L365)                               │
   │  the frontier (L148): the best, the costly (L150)                 │
   └──────────────────────────┬───────────────────────────────────────┘
                              ▼
   ┌────────────────────── THE FINE-TUNED (L365) ───────────────────────┐
   │  the niche (L365): the legal (L371), the format (L143)            │
   └──────────────────────────────────────────────────────────────────┘

The flow is the strategy: route → tier → fine-tuned (L365).

How It Works — The Strategy, Part by Part

  • The tiers (L365). The small (L157), the mid (L365), the frontier (L148) — the cost (L150) and the quality (L341) ladder (L365).
  • The routing (L155). The tier per request (L365): the classifier (L353) — the complexity (L365) — the tier (L155).
  • The fine-tuning (L365). The custom (L365): the domain (L371), the format (L143) — the high-volume (L358) niche (L365).
  • The economics (L334). The tokens (L332) per tier (L365) — the L334 attribution (L334) — the tier's (L365) cost vs the quality (L341).

Note

The routing is the strategy's cost lever (L365). The senior answer routes (L155): the tier per request (L365) — the simple (L365) to the small (L157) — the 80% (L365) of the requests (L358) on the cheap (L150); the hard (L365) to the frontier (L148) — the 20% (L365) on the costly (L150). The routing (L155) — the classifier (L353) with the fallback (L365): the uncertain (L139) → the next tier (L365) — the cost (L334) and the quality (L341) balanced (L365).

Real Project Usage

  • A support copilot (L350). The tiered (L365): the triage (L350) on the small (L157), the resolution (L350) on the mid (L365).
  • A document processor (L353). The extraction (L143) on the fine-tuned (L365) — the high-volume (L358) invoices (L353).
  • A chat product (L348). The routing (L155): the simple (L365) chats on the small (L157), the complex (L365) on the frontier (L148).
  • A legal or medical copilot (L371). The fine-tuned (L365) — the domain (L371) — with the compliance (L371).
  • Anything enterprise (L380). The strategy (L365) — the tiers, the routing, the fine-tuning (L365).

The through-line: the strategy is the model's — the tiers, the routing, and the fine-tuning (L365).

Interview Explanation

Say it in four moves:

  1. The tiers. "The small (L157), the mid (L365), the frontier (L148)."
  2. The routing. "The tier per request — the classifier (L353), the fallback (L365)."
  3. The fine-tuning. "The custom — the domain (L371), the format (L143)."
  4. The economics. "The tokens (L332) per tier (L365) — the cost vs the quality (L341)."

Senior-Level Insights

  • The 80/20 is the routing's (L365). The 80% (L365) of the requests (L358) on the small (L157) — the 20% (L365) on the frontier (L148) — the cost (L334) shaped (L365).
  • The fallback is the quality's (L365). The uncertain (L139) routing (L155) — the next tier (L365) — the quality (L341) preserved (L365).
  • The fine-tuning is the niche's (L365). The high-volume (L358) specific (L365) task (L365) — the legal (L371), the format (L143) — the fine-tuned (L365) model (L148).
  • The drift is the strategy's watch (L335). The model (L148) updates (L335) — the tiers (L365) re-evaluated (L341) — the L335 watch (L335), strategy-shaped (L365).
  • The eval is the tier's (L341). The golden set (L342) per tier (L365) — the L341 suite (L341) gating (L365).

Common Mistakes

  • The one model (L365). The frontier (L148) for everything (L365) — the cost (L334) — the tiers (L365) are the economics (L334).
  • The rule-based routing (L155). The static rules (L155) — the classifier (L353) and the fallback (L365) — the dynamic (L155).
  • The fine-tune-everything (L365). The fine-tuning (L365) for the commodity (L363) — the domain (L371) niche (L365) only (L365).
  • The tier un-evaluated (L341). The tier (L365) without the golden set (L342) — the quality (L341) unknown (L365).
  • The drift un-watched (L335). The model (L148) update (L335) — the tiers (L365) un-re-evaluated (L365).

Best Practices

  • Tier the models (L365) — the small (L157), the mid (L365), the frontier (L148).
  • Route by the complexity (L155) — the classifier (L353), the fallback (L365).
  • Fine-tune the niche (L365) — the high-volume (L358) domain (L371).
  • Eval per tier (L341) — the golden set (L342) gating (L365).
  • Watch the drift (L335) — the re-evaluation (L341) on the update (L365).

Interview Questions

Q: Walk me through the model selection at scale.

A: The strategy (L365). The tiers — the small (L157), the mid (L365), the frontier (L148). The routing — the tier per request (L155). The fine-tuning — the custom for the niche (L365). And the economics — the tokens (L332) per tier (L365).

Q: How do you route?

A: The classifier (L353): the request's (L365) complexity (L365) → the tier (L155) — the simple (L365) to the small (L157), the hard (L365) to the frontier (L148). The fallback (L365): the uncertain (L139) routing (L155) → the next tier (L365) — the quality (L341) preserved (L365).

Q: When do you fine-tune?

A: The niche (L365): the high-volume (L358) specific (L365) task (L365) — the legal (L371) or the medical (L371) domain (L365), the structured format (L143) — the fine-tuned (L365) model (L148) — the quality (L341) and the cost (L334) improved (L365). The commodity (L363) — the general chat (L348) — stays on the API (L278).

Q: What's the economics?

A: The tier's (L365) cost vs the quality (L341): the tokens (L332) per tier (L365) — the 80/20 (L365): the 80% (L365) on the small (L157), the 20% (L365) on the frontier (L148) — the cost (L334) shaped (L365). The L334 attribution (L334) per tier (L365) — the budget (L368) informed (L365).

Follow-Up Questions

  • What are the tiers (L365)?
  • How do you route (L155)?
  • When do you fine-tune (L365)?
  • What's the economics (L334)?
  • What's the drift watch (L335)?

Comparison Table — The Tiers

Tier (L365)The use (L365)The cost (L150)The latency (L333)
The small (L157)the classify (L353), the extract (L143)the lowest (L150)the fastest (L151)
The mid (L365)the summarize (L166)the mid (L365)the mid (L365)
The frontier (L148)the reasoning (L203)the highest (L150)the slowest (L333)

The senior read: the tier matches the task — the ladder (L365).

Code Example — The Strategy, Applied

// The model strategy (L365) — the tiers, the routing, the fine-tuning (L365).
// 1 · THE TIERS (L365) — the ladder (L365).
const tiers = {
  small:    'titan-text-lite',        // the cheap and the fast (L157, L151)
  mid:      'claude-3-5-haiku',       // the balance (L365)
  frontier: 'claude-3-5-sonnet',      // the best and the costly (L148, L150)
  niche:    'fine-tuned-legal',       // the custom (L365, L371)
};

// 2 · THE ROUTER (L155) — the complexity (L365).
async function route(request) {
  const complexity = await classify(request);      // the classifier (L353)
  if (complexity === 'simple') return tiers.small; // L157
  if (complexity === 'mid')    return tiers.mid;   // L365
  if (complexity === 'hard')   return tiers.frontier;  // L148
  // the fallback (L365): the uncertain (L139) → the next tier (L365)
  return tiers.frontier;
}

// 3 · THE FINE-TUNED (L365) — the niche (L365).
if (request.domain === 'legal') return tiers.niche;   // L371, L365

// 4 · THE ECONOMICS (L334) — the tokens (L332) per tier (L365).
await metering.write({ tier, tokens: usage });        // L332, L334

// 5 · THE EVAL (L341) — the golden set (L342) per tier (L365).
//   the L341 suite (L341) gating the tiers (L365)
What the reader must SEE — the strategy, applied:

  the four tiers             → the ladder (L365)
  classify → the tier        → the routing (L155, L353)
  the uncertain → the frontier → the fallback (L365)
  the legal → the fine-tuned → the niche (L365, L371)
  metering per tier          → the economics (L332, L334)

  The tiers, the routing, the fine-tuning (L365).

Tip

The pair that defines the strategy: the complexity router (the tier per request, L155) and the per-tier metering (the economics, L334). Tier the models, route by the complexity, fine-tune the niche, eval per tier — the scale's model strategy (L365).

Performance Notes

  • The small tier is the latency (L157). The fast (L151) — the classify (L353) and the extract (L143) (L365).
  • The routing is the cost (L334). The 80/20 (L365) — the tokens (L332) shaped (L365).
  • The fine-tuned is the niche's (L365). The high-volume (L358) — the tokens (L332) reduced (L365).
  • The drift is the watch's (L335). The updates (L335) — the re-eval (L341) on the schedule (L221).

Debugging Scenarios

SymptomFirst check (L365)The lever
The cost explodesThe routing (L155)The 80/20 (L365)
The quality dipsThe fallback (L365)The uncertain (L139) → the frontier (L148)
The niche is slowThe fine-tuning (L365)The custom model (L365)
The tier regressesThe evals (L341)The golden set (L342)
The model updatedThe drift (L335)The re-eval (L341)

Quick Revision Notes

  • The model selection at scale = the model's strategy (L365): the tiers, the routing, the fine-tuning.
  • The tiers: the small (L157), the mid (L365), the frontier (L148).
  • The routing: the tier per request (L155) — the classifier (L353), the fallback (L365).
  • The fine-tuning: the custom — the domain (L371), the format (L143).
  • The economics: the tokens (L332) per tier (L365) — the cost vs the quality (L341).

Cheat Sheet

MODEL SELECTION AT SCALE = the tiered strategy

THE TIERS (L365)
  the small (L157) — the cheap (L150), the fast (L151):
    the classify (L353), the extract (L143)
  the mid (L365) — the balance: the summarize (L166)
  the frontier (L148) — the best (L341), the costly (L150):
    the reasoning (L203)
  the fine-tuned (L365) — the niche (L365): the domain (L371),
    the format (L143)

THE ROUTING (L155)
  the tier per request (L365) — the classifier (L353)
  the simple (L365) → the small (L157) · the hard (L365) → the frontier (L148)
  the fallback (L365): the uncertain (L139) → the next tier (L365)
  the 80/20 (L365): the 80% on the small (L157)

THE FINE-TUNING (L365)
  the high-volume (L358) specific (L365) task (L365)
  the legal (L371), the medical (L371), the format (L143)

THE ECONOMICS (L334)
  the tokens (L332) per tier (L365) · the L334 attribution (L334)
  the tier's (L365) cost vs the quality (L341)

INTERVIEW, 4 MOVES
  1 tiers    "the small, the mid, the frontier (L365)"
  2 routing  "the tier per request (L155)"
  3 fine-tuning "the niche (L365)"
  4 economics "the tokens per tier (L334)"

Key Takeaways

Recap

  • The model selection at scale is the routing, the fine-tuning, and the tiered model strategy for the enterprise (L365): the tiers (L365), the routing (L155), the fine-tuning (L365), and the economics (L334)
  • The tiers (L365): the small (L157) — the cheap (L150) and the fast (L151); the mid (L365) — the balance; and the frontier (L148) — the best (L341) and the costly (L150)
  • The routing (L155): the tier per request (L365) — the classifier (L353) with the fallback (L365) — the 80/20 (L365)
  • The fine-tuning (L365): the custom (L365) — the domain (L371), the format (L143) — for the high-volume (L358) niche (L365)
  • The economics (L334): the tokens (L332) per tier (L365) — the L334 attribution (L334) — the tier's (L365) cost vs the quality (L341)
  • The AI shape (L365): the enterprise (L380) — the models (L148) tiered (L365), routed (L155), and fine-tuned (L365) — with the evals (L341) per tier (L365) and the drift (L335) watched (L365)

Check your understanding

Answer each one out loud. Then rate yourself honestly.

0/8
1

What are the tiers (L365)?

2

How do you route (L155)?

3

When do you fine-tune (L365)?

4

What's the economics (L334)?

5

What's the drift watch (L335)?

6

What's the fallback (L365)?

7

What's the 80/20 (L365)?

8

What is the tiered strategy (L365)?

A Closing Note — The Firm, Tiered

You now hold the strategy: the tiers, the routing, and the fine-tuning — with the intake routing and the specialists hired. The law firm matches the cases to the partners — and the niches have the specialists (L365).

Next: the AWS vs the Azure vs the GCP for AI, and the exit-cost question — Cloud Selection (L366).