Technology Selection

The technology selection is the frameworks, the databases, and the shortlist you can defend (L362) — the criteria (the fit: the requirements L359 match; the cost: the licensing L368 and the infra L285; the risk: the maturity L362 and the lock-in L377, L362), the shortlist (the candidates: the 3-5 considered L362), and the decision (the ADR L361: the choice with the context L361 and the consequences L361, L362) — the enterprise's (L380) selections, defended (L362).

The one-sentence interview answer: "The technology selection is the defensible choice (L362). The criteria (L362): the fit (L362) — the requirements (L359) match: the latency (L333), the scale (L358), the compliance (L371); the cost (L362) — the licensing (L362), the infra (L285), the team's (L362) skill (L362); and the risk (L362) — the maturity (L362), the community (L362), the lock-in (L377). The shortlist (L362): the candidates (L362) — the 3-5 (L362) considered (L362): the frameworks (L362), the databases (L268, L182), the clouds (L366). The decision (L361): the ADR (L361) — the choice (L361) with the context (L361) and the consequences (L361). The AI's criteria (L362): the model (L148) — the fit and the cost (L334); the framework (L214) — the LangChain (L214) vs the plain (L362); and the vector DB (L182) — the pgvector (L183) vs the Pinecone (L184). The AI shape (L173): the enterprise (L380) — the framework (L362), the vector DB (L182), the cloud (L366) — selected (L362) by the criteria (L362), recorded (L361), and defended (L362)."

Think of the technology selection as the chef's ingredient choice. The chef (the architect, L362) chooses the ingredients (the technologies, L362) for the dish (the system, L173). The criteria (L362): the fit (L362) — the dish's (L359) needs (L362): the flavor (the latency, L333), the portions (the scale, L358); the cost (L362) — the price (the licensing, L362) and the pantry (the infra, L285); and the risk (L362) — the freshness (the maturity, L362) and the supplier (the lock-in, L377). The shortlist (L362): the 3-5 (L362) ingredients (L362) — the tested (L362). The choice (the decision, L361): the ADR (L361) — the ingredient (L361), the why (L361), the trade-offs (L361). The dish works because the criteria are clear, the shortlist is tested, and the choice is recorded (L362).

   the ingredient choice (the selection, L362)
   ┌────────────────────────────────────────────────────────┐
   │ the criteria (L362) — the fit, the cost, the risk      │
   │ the shortlist (L362) — the 3-5 candidates (L362)       │
   │ the choice (the ADR, L361) — the why and the           │
   │ trade-offs (L361)                                      │
   └────────────────────────────────────────────────────────┘

The mental model is the ingredient choice: the criteria, the shortlist, and the recorded choice (L362).

   the requirement (L359)
        │  the vector store (L182) needed
        ▼
   ┌────────────────────── THE CRITERIA (L362) ─────────────────────────┐
   │  the fit (L362): the scale (L358), the latency (L333)             │
   │  the cost (L362): the licensing (L362), the infra (L285)          │
   │  the risk (L362): the maturity (L362), the lock-in (L377)         │
   └──────────────────────────┬───────────────────────────────────────┘
                              ▼
   ┌────────────────────── THE SHORTLIST (L362) ────────────────────────┐
   │  the pgvector (L183) · the Pinecone (L184) · the Qdrant (L185)    │
   └──────────────────────────┬───────────────────────────────────────┘
                              ▼
   ┌────────────────────── THE DECISION (L361) ─────────────────────────┐
   │  ADR-021: the pgvector (L183) — the fit (L183), the cost (L362),  │
   │  the operations (L268) — the choice, recorded (L361)              │
   └──────────────────────────────────────────────────────────────────┘

The flow is the selection: criteria → shortlist → decision (L362).

• The criteria (L362). The fit (L362), the cost (L362), the risk (L362) — the requirements (L359) match (L362).
• The shortlist (L362). The candidates (L362) — the 3-5 (L362) considered (L362).
• The decision (L361). The ADR (L361) — the choice (L361) with the context (L361) and the consequences (L361).
• The AI's criteria (L362). The model (L148), the framework (L214), the vector DB (L182).

• An enterprise engagement (L379). The framework (L362), the database (L268), the cloud (L366) — selected (L362) and recorded (L361).
• A vector DB choice (L182). The pgvector (L183) vs the Pinecone (L184) — the criteria (L362) and the ADR (L361).
• A framework choice (L362). The LangChain (L214) vs the plain SDK (L152) — the fit and the cost (L362).
• A cloud choice (L366). The AWS (L366) vs the Azure (L366) — the exit cost (L366).
• Anything enterprise (L380). The selection (L362) — the criteria, the shortlist, the ADR (L362).

The through-line: the method is the selection's — the criteria, the shortlist, and the decision (L362).

Say it in four moves:

1. The criteria. "The fit, the cost, the risk (L362)."
2. The shortlist. "The 3-5 candidates (L362)."
3. The decision. "The ADR (L361) — the choice with the why (L361)."
4. The AI's. "The model (L148), the framework (L214), the vector DB (L182)."

• The fit is the requirements' (L362). The latency (L333), the scale (L358), the compliance (L371) — the requirements (L359) decide (L362).
• The cost is the total's (L362). The licensing (L362), the infra (L285), the team (L362) — the L368 total cost (L368) (L362).
• The risk is the lock-in's (L377). The maturity (L362) and the exit cost (L366) — the L377 lock-in (L377), selection-shaped (L362).
• The ADR is the defense (L361). The choice (L361) with the context (L361) — the selection (L362) defended (L362).
• The AI's vector DB is the RAG's (L182). The pgvector (L183) vs the Pinecone (L184) — the scale (L358) and the ops (L268) decide (L362).

• The favorite picked (L362). The familiar (L362) without the criteria (L362) — the fit (L362) un-checked (L362).
• The single option (L362). The one candidate (L362) — the shortlist (L362) of the 3-5 (L362).
• The cost-blind (L362). The licensing (L362) and the infra (L285) un-estimated (L362) — the L368 budget (L368) blown (L362).
• The lock-in ignored (L377). The proprietary (L362) without the exit (L366) — the L377 cost (L377) (L362).
• The un-recorded choice (L361). The decision (L362) without the ADR (L361) — the defense (L362) lost (L361).

• Score by the criteria (L362) — the fit, the cost, the risk (L362).
• Shortlist the 3-5 (L362) — the candidates (L362).
• Estimate the total cost (L368) — the licensing (L362), the infra (L285).
• Check the lock-in (L377) — the exit cost (L366).
• Record the ADR (L361) — the selection (L362) defended (L362).

Q: Walk me through the technology selection.

A: The defensible choice (L362). The criteria — the fit, the cost, the risk (L362). The shortlist — the 3-5 candidates (L362). The decision — the ADR (L361). And the AI's — the model (L148), the framework (L214), the vector DB (L182).

Q: How do you pick the vector DB?

A: The criteria (L362): the fit — the scale (L358) and the latency (L333) of the retrieval (L189); the cost — the licensing (L362) and the infra (L285); and the risk — the maturity (L362) and the ops (L268). The shortlist (L362): the pgvector (L183), the Pinecone (L184), the Qdrant (L185). The ADR (L361) records the choice (L361).

Q: How do you score the candidates?

A: The weighted criteria (L362): the fit (L362) — the requirements (L359) match (L362); the cost (L362) — the licensing (L362) and the infra (L285); the risk (L362) — the maturity (L362) and the lock-in (L377). Each candidate (L362) scored (L362) — the table (L362) in the ADR (L361).

Q: What's the AI-specific selection?

A: Three (L362): the model (L148) — the fit and the cost (L334); the framework (L214) — the LangChain (L214) vs the plain SDK (L152); and the vector DB (L182) — the pgvector (L183) vs the managed (L184). Each (L362) by the criteria (L362), recorded (L361).

• What are the criteria (L362)?
• How do you pick the vector DB (L182)?
• How do you score the candidates (L362)?
• What's the AI-specific selection (L362)?
• What's the ADR (L361)?

The senior read: the three axes, scored and recorded (L362).

// The technology selection (L362) — the criteria, the shortlist, the ADR (L362).
// 1 · THE CRITERIA (L362) — the weighted axes (L362).
const criteria = {
  fit:   { weight: 0.5, score: (c) => fitScore(c, requirements) },  // L359
  cost:  { weight: 0.3, score: (c) => costScore(c) },               // L368
  risk:  { weight: 0.2, score: (c) => riskScore(c) },               // L377
};

// 2 · THE SHORTLIST (L362) — the candidates (L362).
const shortlist = [
  { id: 'pgvector',  fit: 9, cost: 9, risk: 8 },   // L183
  { id: 'pinecone',  fit: 8, cost: 6, risk: 7 },   // L184
  { id: 'qdrant',    fit: 7, cost: 7, risk: 7 },   // L185
];

// 3 · THE SCORE (L362) — the weighted sum (L362).
function score(c) {
  return criteria.fit.weight * c.fit
       + criteria.cost.weight * c.cost
       + criteria.risk.weight * c.risk;
}
const ranked = shortlist.map((c) => ({ ...c, total: score(c) }))
                       .sort((a, b) => b.total - a.total);
// pgvector wins (L362)

// 4 · THE ADR (L361) — the choice recorded (L361).
//   ADR-021: the pgvector (L183) — the fit, the cost, the ops (L361)

What the reader must SEE — the method, applied:

  fit 0.5 + cost 0.3 + risk 0.2 → the criteria (L362)
  the 3-candidate shortlist     → the options (L362)
  the weighted total            → the score (L362)
  pgvector wins → the ADR       → the record (L361)

  The criteria, the shortlist, the decision (L362).

• The selection is the project's speed (L362). The decided stack (L362) — the rework (L362) avoided (L362).
• The cost is the budget's (L362). The licensing (L362) and the infra (L285) — the L368 estimate (L368) (L362).
• The lock-in is the exit's (L377). The proprietary (L362) — the L377 exit cost (L377) (L362).
• The ADR is the review's (L361). The context (L361) — the selection (L362) revisited (L361).

• The technology selection = the defensible choice (L362): the criteria, the shortlist, the decision.
• The criteria: the fit, the cost, the risk (L362).
• The shortlist: the 3-5 candidates (L362).
• The decision: the ADR (L361) — the choice with the why (L361).
• The AI's: the model (L148), the framework (L214), the vector DB (L182).

TECHNOLOGY SELECTION = the shortlist you can defend

THE CRITERIA (L362)
  the fit (L362) — the requirements (L359) match:
    the latency (L333), the scale (L358), the compliance (L371)
  the cost (L362) — the licensing (L362), the infra (L285),
    the team (L362)
  the risk (L362) — the maturity (L362), the community (L362),
    the lock-in (L377)

THE SHORTLIST (L362)
  the 3-5 candidates (L362) — the frameworks (L362),
  the databases (L268, L182), the clouds (L366)
  the weighted score (L362)

THE DECISION (L361)
  the ADR (L361) — the choice (L361), the context (L361),
  the consequences (L361)

THE AI'S (L362)
  the model (L148) — the fit and the cost (L334)
  the framework (L214) — the LangChain (L214) vs the plain (L152)
  the vector DB (L182) — the pgvector (L183) vs the Pinecone (L184)

INTERVIEW, 4 MOVES
  1 criteria  "the fit, the cost, the risk (L362)"
  2 shortlist "the 3-5 candidates (L362)"
  3 decision  "the ADR (L361)"
  4 the AI's  "the model, the framework, the vector DB (L362)"

You now hold the method: the criteria, the shortlist, and the decision — with the fit scored and the choice recorded. The chef chose by the dish — and the recipe is in the logbook (L362).

Next: the managed APIs, the open-source stacks, or the in-house — Build vs Buy (L363).