Module 2 Assignment: Embeddings and semantic similarity

Scenario

You are advising a product team evaluating an NLP workflow before using it in customer-facing communication. The stakeholders are: product manager, support lead, privacy reviewer, and model evaluator.

Task

Answer the module question: How do vector representations support semantic operations?

Use the module lab and course readings to produce: NLP evaluation packet with task framing, retrieval/evaluation design, and deployment guardrails focused on embeddings and semantic similarity: Build a simple semantic search over course documents..

Required Evidence

• Define the decision or system boundary in one paragraph.

• Identify the dataset, proxy data, or evidence source you used: synthetic support messages, retrieval snippets, intent labels, and factuality checks.

• Compare at least two alternatives, baselines, policies, or designs.

• Report one quantitative result or structured scoring table.

• Explain two failure modes and one mitigation for each.

• State what additional evidence would be required before real deployment.

Submission

Submit the completed notebook plus a 900-1200 word memo. The memo must include clear headings for context, method, evidence, risks, recommendation, and open questions.

# Assignment workspace for Module 2: Embeddings and semantic similarity
module = 2
decision = "How do vector representations support semantic operations?"
artifact = "NLP evaluation packet with task framing, retrieval/evaluation design, and deployment guardrails focused on embeddings and semantic similarity: Build a simple semantic search over course documents."

alternatives = [
    {"option": "baseline_or_manual_process", "strength": "", "risk": "", "evidence": ""},
    {"option": "ai_assisted_or_advanced_option", "strength": "", "risk": "", "evidence": ""},
]

recommendation = {
    "decision": decision,
    "recommended_option": "",
    "minimum_evidence_before_pilot": [],
    "monitoring_metric": "",
    "rollback_trigger": "",
}

{"module": module, "artifact": artifact, "alternatives": alternatives, "recommendation": recommendation}

{'module': 2,
 'artifact': 'NLP evaluation packet with task framing, retrieval/evaluation design, and deployment guardrails focused on embeddings and semantic similarity: Build a simple semantic search over course documents.',
 'alternatives': [{'option': 'baseline_or_manual_process',
   'strength': '',
   'risk': '',
   'evidence': ''},
  {'option': 'ai_assisted_or_advanced_option',
   'strength': '',
   'risk': '',
   'evidence': ''}],
 'recommendation': {'decision': 'How do vector representations support semantic operations?',
  'recommended_option': '',
  'minimum_evidence_before_pilot': [],
  'monitoring_metric': '',
  'rollback_trigger': ''}}

Acceptance Criteria

Your submission is complete only if another reviewer can reproduce your reasoning from the evidence you provide. You do not need production-grade data, but you must be explicit about proxy-data limits and what would change with real institutional data.