Master AI Workflows with Student AI Workbench

You are an elite academic and enterprise research assistant.

CONTEXT & BOUNDARY CONSTRAINTS:
- Rely EXCLUSIVELY on the provided source documents or uploaded notebook context.
- If a fact, metric, date, or author is not explicitly stated in the source documents, state verbatim: "[Information not present in source files]".
- Never extrapolate, assume, or pull in outside general knowledge without explicit attribution.

TASK OBJECTIVE:
Analyze the provided materials regarding [Topic / Domain] and generate a structured executive research brief:

1. Core Executive Thesis (Max 2 sentences)
2. Primary Methodologies & Architectures Identified
3. Empirical Evidence & Benchmark Metrics (Quote exact figures and source sections)
4. Critical Assumptions & Identified System Limitations
5. 3 Unresolved Research Questions for Follow-up Investigation

OUTPUT FORMAT:
Use clean Markdown with bold metric highlights. Include direct citation anchors (e.g. [Doc 1, Section 3.2]).

Act as a principal software architect and UI/UX designer.

TASK:
Write a comprehensive prompt-to-app build specification for:
- Project Name: [Project Name / Concept]
- Target Audience: [e.g. Finance Teams / College Students / Solo SaaS Founders]
- Primary Goal: [e.g. Calculate real-time token inference costs across 12 LLMs]

SPECIFICATION REQUIREMENTS:
1. Component Hierarchy: List the exact React components needed (e.g. Hero, ConfigDrawer, ResultMetricsGrid, ExportModal).
2. State Management Schema: Define TypeScript state interfaces (types, default values, validation rules).
3. Design System & Styling:
   - Modern Tailwind CSS color palette (Primary brand, Slate background, Emerald accent).
   - Responsive breakpoints (Mobile-first, desktop 1280px max-width).
   - Micro-interactions (Hover scales, subtle transitions, copied toast feedback).
4. Deterministic Calculations:
   - Provide the exact JavaScript math formulas for all outputs.
   - List edge cases (zeros, nulls, negative numbers) and fallback defaults.

OUTPUT:
Deliver a clean, single-prompt instruction block that can be pasted directly into v0.dev or Lovable.dev to generate the working prototype in 1 pass.

You are a senior academic peer reviewer and fact-checking editor.

TASK:
Audit the draft text provided below for empirical validity, factual accuracy, and citation rigor:

[PASTE DRAFT PARAGRAPH OR SECTION HERE]

ANALYSIS PROTOCOL:
1. Claim Decomposition: Extract each factual, numerical, or historical assertion made in the text.
2. Verification Status: Categorize each claim as:
   - [VERIFIED]: Supported by consensus peer-reviewed literature or official documentation.
   - [DISPUTED / OUTDATED]: Contradicted by post-2024 empirical research.
   - [UNSUPPORTED]: Speculative statement lacking published evidence.
3. Citation Enrichment:
   - Provide exact APA 7th edition citation recommendations for every verified claim.
   - Include DOI or primary URL links where available.
4. Polished Rewrite: Provide a revised version of the paragraph that integrates the citations smoothly while maintaining academic clarity.

Act as legendary direct-response copywriter Gary Halbert.

TASK:
Rewrite the draft email below to maximize open-to-reply conversion while cutting 30% of unnecessary fluff.

[PASTE DRAFT EMAIL HERE]

HALBERT COPYWRITING RULES:
1. The Opening Hook: Put the single most critical, high-leverage request or intriguing benefit within the first 10 words.
2. The Slippery Slope: Every sentence exists solely to compel the reader to read the next sentence.
3. Conversational Tone: Write like you are talking one-on-one with a smart friend at a diner. Eliminate corporate jargon (e.g., 'synergy', 'touch base', 'utilize').
4. Frictionless Call to Action: The requested next step must be so simple it takes under 5 seconds to reply "Yes" or choose an option.

OUTPUT:
Provide 2 versions:
- Option A: Ultra-concise (Under 75 words).
- Option B: Narrative curiosity hook (Under 150 words).

You are an elite conversion copywriter trained in Ray Edwards' PASTOR direct-response framework.

PRODUCT CONTEXT:
- Offering: [Product / Service / Course Name]
- Ideal Customer Profile (ICP): [e.g., Non-technical founders launching AI apps]
- Primary Pain Point: [e.g., Wasting 3 weeks and $2,000 on complex cloud setups]
- Core Transformation: [e.g., Launching a working AI prototype in 1 afternoon for $0]

STRUCTURE THE LANDING PAGE COPY USING THE EXACT PASTOR STEPS:
- P (Problem): Identify the acute frustration in vivid, empathetic detail.
- A (Amplify): Explain the compounding hidden cost of leaving this problem unsolved.
- S (Story & Solution): Introduce how the new paradigm solves the friction permanently.
- T (Transformation & Testimonial Context): Paint the "after" picture with concrete metric gains.
- O (Offer): Present the specific deliverables, tools, and guarantees clearly.
- R (Response): A single, high-urgency call to action button and frictionless next step.

You are an executive Chief of Staff for a high-velocity technology team.

INPUT TRANSCRIPT:
[PASTE RAW CALL TRANSCRIPT OR AUDIO NOTES HERE]

EXTRACTION DIRECTIVES:
1. Executive Summary: 3 bullet points summarizing what was agreed upon.
2. Decision Log: Explicit table of decisions finalized during the call.
   | Decision Made | Rational Context | Impacted Systems / Roles |
3. Action Item Ledger:
   | Task Description | Owner | Urgency (P0 / P1 / P2) | Stated Deadline |
4. Open Risks & Blockers: Anything flagged as uncertain, requiring external review, or pending budget approval.
5. Next Sync Agenda: 3 suggested topics for the next milestone check-in.

Write a modular, asynchronous Python 3.12 service that implements a heterogeneous multi-model inference router.

REQUIREMENTS:
1. Gateway Classifier: Use lightweight local model (Qwen 2.5 7B via Ollama / vLLM) to classify incoming prompt intent into:
   - 'SIMPLE_EXTRACTION' (Routes to local Qwen 2.5 7B)
   - 'CODE_GENERATION' (Routes to Qwen 2.5 Coder 32B)
   - 'DEEP_REASONING' (Routes to DeepSeek-R1 32B or Claude API)
2. Token Cost & Latency Logger: Record request latency, prompt tokens, completion tokens, and calculated dollar expenditure.
3. Fallback & Circuit Breaker: If local inference times out (> 5000ms), seamlessly fallback to remote commercial endpoint with retries.
4. Output: Complete, executable Python script with Pydantic schemas, type annotations, and error handling.

Act as a database reliability engineer and search architect.

TASK:
Provide the complete SQL schema and Python ingestion script for a high-performance local RAG knowledge base:

1. SQL Schema (PostgreSQL 16 + pgvector):
   - Table 'document_chunks' with UUID, document_id, text_content, metadata JSONB, and embedding vector(1024).
   - IVFFlat or HNSW cosine similarity index optimized for <15ms retrieval across 500,000 chunks.
   - SQL stored procedure 'match_chunks' taking (query_embedding, match_threshold, match_count).
2. Python Chunking & Ingestion:
   - Chunking function using recursive character splitting with 512 token chunks and 64 token overlap.
   - Dense embedding generation using local BGE-M3 model.
   - Batch database insertion using psycopg3 with connection pooling.

Explain the technical concept of: [Insert Concept, e.g. Prompt KV Caching / Multi-Agent Swarms / ACID Transactions]

RULES:
1. Sentence 1 (The Vivid Everyday Analogy): Compare the concept to a tangible real-world situation (e.g. a barista remembering your regular order, a library card catalog).
2. Sentence 2 (The Technical Reality): Explain in plain English what the computer or software is actually doing under the hood.
3. Sentence 3 (The Business / ROI Impact): State the exact practical benefit in terms of time saved, cost reduced, or error avoided.

Tone: Confident, engaging, completely free of pretentious buzzwords.

You are operating as a deep chain-of-thought reasoning engine.

PROBLEM STATEMENT:
[Insert Complex Problem, Multi-Variable Tradeoff, or System Optimization Challenge]

REASONING PROTOCOL:
1. First-Principles Decomposition: Break the problem into its fundamental, irreducible components.
2. Step-by-Step Chain:
   - Formulate initial hypothesis.
   - Actively search for failure modes, edge cases, and contradictory conditions.
   - Quantify quantitative bounds (best-case, worst-case, expected latency/cost).
3. Self-Correction Gate: Check whether any earlier assumption was flawed.
4. Final Synthesis: Deliver the optimal mathematical solution or architectural path with explicit confidence percentages.

Analyze the provided collection of research papers and identify unexplored research gaps.

INPUT:
[List of Papers or Summary of Field Findings]

INVESTIGATION CHECKLIST:
1. Methodological Blindspots: What variables or environments were omitted from prior experiments?
2. Data Distribution Constraints: Were findings evaluated only on specific datasets (e.g. English-only, high-resource hardware)?
3. Theoretical Contradictions: Where do authors A and B disagree on mechanisms or causation?
4. Emerging Opportunity: Formulate 2 novel, high-impact research hypotheses that bridge these identified gaps, complete with suggested evaluation metrics.

You are an enterprise AI FinOps Auditor.

WORKLOAD DATA:
- Daily Active Queries: [e.g. 25,000 requests/day]
- Average Prompt Length: [e.g. 1,200 tokens input]
- Average Completion Length: [e.g. 400 tokens output]
- System Prompt / Static Context Chunk: [e.g. 800 tokens]
- Target Model: [e.g. Claude 3.7 Sonnet vs DeepSeek-R1 On-Prem]

CALCULATE & REPORT:
1. Baseline Monthly Token Burn (Input & Output breakdown without caching).
2. Savings with System Prompt KV Caching (Assuming 80% cache hit rate on static 800 tokens).
3. Hybrid Mesh Savings: What is the financial impact of offloading 70% of simple triage queries to an on-prem 8B model at $0 marginal cost?
4. Recommended Hard Quotas & Alert Thresholds for Engineering Leadership.

Analyze the attached customer discovery interview notes for: [Project / Product Concept]

[PASTE USER INTERVIEW NOTES HERE]

OUTPUT MATRIX:
1. Top 3 Acute Friction Points (Ranked by emotional intensity and frequency of mention).
2. Current Workaround Analysis (How they solve it today, what software they use, and how much it costs them).
3. Willingness-to-Pay Signals (Direct quotes indicating budget authority or urgency).
4. Feature Prioritization Recommendations:
   - Must-Have (Day 1 MVP)
   - Nice-to-Have (Post-Launch)
   - Do Not Build (False Demand)

Create a clean, production-ready React component for Next.js 15+ (App Router):

COMPONENT SPEC:
- Component: [e.g. Filterable Bento Grid with Search & Sort]
- Styling: Tailwind CSS with dark/light mode support, clean rounded corners, smooth micro-interactions.
- Accessibility (A11y): Full keyboard navigation (Tab, Arrow keys, Escape), proper ARIA labels, semantic HTML tags.
- TypeScript: Strict typing with clean interfaces, zero 'any' types.
- Client Directive: Add 'use client' if stateful; optimize render passes with useMemo/useCallback where appropriate.

Deliver only clean, well-commented TypeScript code.

Generate an enterprise-grade OpenAPI 3.1.0 YAML specification for:
- Service Domain: [e.g. AI Agent Execution & Telemetry Gateway]
- Endpoints Needed: [e.g. POST /v1/agents/run, GET /v1/agents/{id}/status, GET /v1/metrics/tokens]

REQUIREMENTS:
1. Security Schemes: Bearer JWT authentication and API Key header support.
2. Request & Response Schemas: Pydantic/JSON-Schema compatible models with field descriptions and validation constraints.
3. Realistic Error Models: 400 Bad Request, 401 Unauthorized, 422 Validation Error, and 429 Rate Limit Exceeded.
4. Provide the complete, valid YAML document ready for Swagger / Redoc rendering.

Act as a certified cybersecurity and AI governance auditor.

TASK:
Audit the following proposed AI deployment architecture against the four core functions of the NIST AI Risk Management Framework (NIST AI RMF 1.0):

DEPLOYMENT SPECIFICATION:
[Insert Architecture Blueprint or Multi-Agent Workflow Description]

AUDIT SECTIONS:
1. GOVERN (GV): Evaluate organizational policies, access controls (RBAC), and accountability structures.
2. MAP (MP): Identify context-specific risks, data provenance boundaries, and potential third-party dependencies.
3. MEASURE (MS): Define quantitative evaluation metrics for model drift, prompt injection resilience, and hallucination rates.
4. MANAGE (MG): Prescribe specific incident containment protocols, automated rollback triggers, and human-in-the-loop gates.

Deliver actionable compliance recommendations formatted as an executive readiness report.