Prompt Cards

Design a hybrid retrieval strategy combining vector search, lexical search, and a knowledge graph layer (‘Graph-RAG’). Specify when to use which, how to evaluate, and how to keep citations reliable.

Create an eval-first LLMOps design: golden sets, adversarial tests, continuous regression, cost/latency tracking, and release gates. Include a ‘model change control’ policy.

Build a P&L-style ROI model for deploying agents as digital workers: labor displacement vs augmentation, quality risk, error cost, cycle time reduction, and governance overhead. Provide a template with assumptions.

Create a physical AI roadmap for 2026: use cases, safety constraints, simulation strategy, edge compute needs, and incident handling. Include a phased pilot plan with success criteria.

Design a preemptive cybersecurity approach: AI-driven exposure management, automated patch prioritization, agent-assisted triage, and ‘human-in-the-loop’ escalation. Include KPIs that reduce mean time to contain.

Write requirements for an AI security platform: prompt/tool injection defenses, model access control, secrets handling, audit trails, red-team workflows, and runtime policy enforcement for agents.

Define a digital provenance program: signing, watermarking policies, asset lineage, approval workflows, and incident response for deepfake risk. Provide a policy template and metric dashboard.

Create a geopatriation plan: which data/models stay in which regions, cross-border processing rules, vendor selection, and operational playbooks. Include risk register and audit evidence checklist.

Design a reference architecture using confidential computing: what runs in TEEs, key management, attestation flows, data minimization, and performance tradeoffs. Include threat model and rollout plan.

Compare building vs renting an AI supercomputing platform: capacity planning, capex/opex, utilization, networking, storage, and governance. Provide a 3-scenario cost curve and decision criteria.

Draft a business case for training/fine-tuning a DSLM: scope, data sources, compliance, expected accuracy lift, cost model vs general LLMs, and rollout phases. Include evaluation and drift monitoring.

Create an adoption plan for an AI-native development platform: target workflows (PRs, tests, docs), guardrails, secure-by-default templates, and developer enablement. Include a ‘tiny teams’ productivity metric model.