Playbook

WhatsApp is where many customers want to talk to businesses, and its rules are strict. This playbook covers platform setup, opt-in, identity verification, grounded answers, confirmed actions, handoff, and evaluation.

SharePoint holds decades of documents behind a complex permission model. AI search over it succeeds only with permission-trimmed indexing, curation, hybrid retrieval, and grounded answers. This playbook covers the build.

Agents act, so evaluating them means evaluating trajectories in realistic environments, not just final text. This playbook covers task suites, simulated environments, trajectory scoring, safety cases, CI gates, and reporting.

Agentic RAG lets the system plan, retrieve iteratively, choose sources, and verify before answering. This playbook covers when it helps, planning, iterative retrieval, verification, budgets, synthesis, and evaluation.

When an AI system is questioned, the audit trail is the answer. This playbook covers the event model, capture points, immutability, redaction, retention, reconstruction, access, and regulatory alignment.

Compliance teams cannot read every communication, transaction, and change. An AI compliance monitor watches defined obligations, flags with evidence, and routes to reviewers. This playbook covers the build.

AI content at scale is only an asset when it is grounded, on-brand, compliant, and reviewed. This playbook covers building a content pipeline with briefs, drafting, validation, editorial gates, and evaluation.

AI spend becomes manageable when it is captured per request, attributed to features and teams, and turned into unit economics. This playbook covers capture, attribution, unit metrics, budgets, alerts, and finance integration.

Prompts are production configuration and deserve the same discipline as code: versioning, review, testing, staged rollout, and rollback. This playbook covers building a prompt management system that provides it.

Some decisions cannot wait for a batch job: fraud, outages, safety events, and customer moments. This playbook covers building a real-time AI monitoring system from streaming ingestion to scoring, alerting, feedback, and evaluation.

A recommendation API has to be relevant, fast, and measurable in the same request. This playbook covers signals, candidate generation, ranking, low-latency serving, experimentation, and monitoring.

Salesforce holds the customer record and a rich permission model that an AI agent must inherit. This playbook covers permissions, data access, actions with approvals, integration with flows, governance, and evaluation.

Semantic search finds meaning, not just keywords, but only when it is combined with lexical search, reranking, and honest relevance evaluation. This playbook covers the build from indexing to operations.

Slack is where work already happens, which makes it the natural home for an AI assistant if it respects permissions, stays grounded, and acts only with approval. This playbook covers the build from app setup to evaluation.

Transcription quality decides what every downstream AI feature can do with voice. This playbook covers audio capture, model selection, diarization, custom vocabulary, post-processing, privacy controls, and evaluation.

A vector search service is shared infrastructure for RAG, semantic search, and recommendations. This playbook covers the embedding pipeline, index choice, filtering, multi-tenancy, latency, freshness, and evaluation.

Approval gates and exception queues are only real if humans can review fast with context. This playbook covers routing policies, reviewer interface, decision capture, staffing, feedback loops, and metrics.

Hybrid search combines exact-term precision with semantic recall and consistently beats either alone on enterprise queries. This playbook covers dual indexes, query processing, fusion, reranking, filters, and evaluation.

A knowledge base chatbot is the most common first LLM project and the most commonly under-built. This playbook covers curation, retrieval, grounded answers, refusal, evaluation, and launch so it earns trust.

Lead scoring works when sales trusts the score and acts on it. This playbook covers outcome definition, fit and intent features, calibrated modeling, CRM delivery, feedback loops, and evaluation on pipeline outcomes.

A model registry is the system of record for what is in production, how it was evaluated, and who approved it. This playbook covers metadata, versioning, lineage, evaluation records, approval, deployment tracking, and access.

Multi-agent systems fail from handoffs, loops, and untraceable behavior. This playbook covers when to decompose, typed contracts, deterministic orchestration, shared state, budgets, trajectory evaluation, and observability.

Some data cannot leave the perimeter, and some workloads justify owning the model. This playbook covers building a private LLM deployment: model selection, serving, security, gateway integration, evaluation, and operations.

Procurement is policy-heavy, approval-heavy, and full of requests that arrive as free text. This playbook covers intake, policy checks, supplier and catalog grounding, approvals, ERP integration, and evaluation.