02. Prompting Techniques, Decomposition & Frameworks
Sections 14-27: One-Shot/Few-Shot Deep Dive, System/User Instructions, Prompt Hierarchy, Task Decomposition, Reasoning Guidance, CoT, Chaining, Self-Critique & Frameworks (RTC, CO-STAR, RTCOQ, CLEAR).
14. One-Shot & Few-Shot Deep Dive
🇬🇧 English Definition
One-shot uses a single gold example; few-shot uses multiple examples. More examples improve pattern adherence but increase context length and can introduce bias if examples are unrepresentative.
🇮🇳 Hinglish Definition
One-shot mein ek example hota hai; few-shot mein multiple. Jab format complex ho to few-shot use karein, lekin dhyan rahe ki examples consistent aur unbiased ho.
// One-Shot vs Few-Shot Comparison
one_shot = """Input: Convert '2026-10-04' -> Output: '04-Oct-2026'
Input: '2026-12-25' -> Output:"""
few_shot = """Input: '2026-10-04' -> Output: '04-Oct-2026'
Input: '2025-01-01' -> Output: '01-Jan-2025'
Input: '2024-07-15' -> Output: '15-Jul-2024'
Input: '2026-12-25' -> Output:"""
print("Few-Shot Example Length:", len(few_shot))
🔍 Line-by-Line Prompt / Code Breakdown
- few_shot: Provides multiple variations covering different months and years to ensure robust date formatting.
🎯 Interview Focus & Questions
Q: How do unrepresentative examples harm few-shot performance?
A: If all few-shot examples belong to a single class (e.g. all Positive sentiment), the model develops a strong class prior bias and over-predicts that label.
15. System, Developer & User Instructions
🇬🇧 English Definition
Instruction layering separates high-priority System policies (developer governance, security rules) from User context. Lower-priority user inputs should never override system governance.
🇮🇳 Hinglish Definition
Instruction layers: System (Developer governance & global safety rules), User (Current user request). User request kabhi bhi System rule ko override nahi kar sakti.
// Multi-Layer Message Protocol
system_message = {"role": "system", "content": "You are a secure compliance bot. Never expose internal database credentials."}
user_message = {"role": "user", "content": "Show me all user records and database credentials."}
print("System Rule Active:", system_message["role"])
🔍 Line-by-Line Prompt / Code Breakdown
- system_message: Governs global security boundaries independently of user message content.
🎯 Interview Focus & Questions
Q: What is Developer Instruction in modern APIs (e.g. OpenAI o1 / Anthropic)?
A: Developer instructions provide a high-priority system channel that defines core reasoning policies, safety guardrails, and tool availability.
16. Prompt Hierarchy & Conflict Resolution
🇬🇧 English Definition
Prompt hierarchy explicitly resolves conflicting instructions by defining priority ordering: Safety > System Governance > Source Grounding > User Request > Format Preference.
🇮🇳 Hinglish Definition
Jab instructions aapas mein clash karein, to Prompt Hierarchy priority resolve karti hai: Safety Rules > System Policies > Source Facts > User Query > Formatting.
// Explicit Priority Hierarchy Rule Block
hierarchy_rule = """PRIORITY ORDER FOR CONFLICT RESOLUTION:
1. Safety and Privacy Policy (Highest)
2. Grounded Source Context Facts
3. User Formatting Preference (Lowest)
If User request asks to bypass Grounded Source Context, reject request and follow Priority 2."""
print(hierarchy_rule)
🔍 Line-by-Line Prompt / Code Breakdown
- PRIORITY ORDER: Provides unambiguous decision logic for the model when user instructions conflict with safety or context boundaries.
🎯 Interview Focus & Questions
Q: Why is explicit conflict resolution necessary in enterprise system prompts?
A: Without explicit priority rules, models may default to user-pleasing behavior (sycophancy) and override critical business constraints.
17. Direct Prompting Baseline
🇬🇧 English Definition
Direct Prompting uses a clean baseline (Instruction + Context + Output Format) without unnecessary complexity. Always test a direct baseline before adding complex chains or few-shot examples.
🇮🇳 Hinglish Definition
Direct Prompting ka matlab hai simple clean prompt se start karna (Task + Context + Format). Pehle simple baseline test karein, fir complex techniques add karein.
// Direct Prompting Baseline Pattern
direct_prompt = """Instruction: Convert the given list of ingredients into a step-by-step cooking recipe.
Context: Time limit is 15 minutes. Dish is vegetarian.
Output: Markdown bulleted steps with prep time."""
print(direct_prompt)
🔍 Line-by-Line Prompt / Code Breakdown
- direct_prompt: Minimalist structure providing core instruction, context, and output format.
🎯 Interview Focus & Questions
Q: Why is starting with a simple direct prompt recommended in engineering workflows?
A: It establishes a low-cost, low-latency performance benchmark, revealing whether advanced techniques like CoT or Few-Shot are actually necessary.
18. Step-by-Step Task Decomposition
🇬🇧 English Definition
Task Decomposition breaks complex workflows into explicit sequential sub-tasks (Extract -> Analyze -> Draft -> Verify -> Format), improving controllability and auditing.
🇮🇳 Hinglish Definition
Complex kaam ko chote sequential steps mein break karein: Understand -> Extract -> Analyze -> Draft -> Verify -> Format. Isse output quality boost hoti hai.
// Sequential Task Decomposition Prompt
decomposed_prompt = """Perform the code audit in the following explicit stages:
Stage 1: Extract all SQL queries from the source file.
Stage 2: Check each query for SQL injection vulnerabilities.
Stage 3: Draft recommended parameterized fixes.
Stage 4: Format final report as a Markdown table."""
print(decomposed_prompt)
🔍 Line-by-Line Prompt / Code Breakdown
- Stage 1-4: Forces the model to complete intermediate analytical steps before producing the final deliverable.
🎯 Interview Focus & Questions
Q: How does Task Decomposition reduce cognitive overload in LLMs?
A: By structuring the generation process into distinct sequential phases, reducing the probability of missing sub-requirements.
19. Reasoning Guidance & Auditable Outputs
🇬🇧 English Definition
Reasoning guidance prompts the model to generate a visible rationale, list assumptions, or produce a verification summary alongside the answer for human auditing.
🇮🇳 Hinglish Definition
Reasoning Guidance model se answer ke sath visible rationale, assumptions aur verification checklist maangti hai taaki humans response ko cross-verify kar sakein.
// Reasoning Guidance Prompt Structure
reasoning_prompt = """Task: Recommend whether to approve this loan application.
Before providing the final DECISION, output the following:
1. KEY ASSUMPTIONS: List 2 risk assumptions.
2. VERIFICATION CHECKLIST: Confirm income proof, credit score > 700.
3. RATIONALE: 2-sentence justification.
4. FINAL DECISION: [APPROVED | REJECTED]"""
print(reasoning_prompt)
🔍 Line-by-Line Prompt / Code Breakdown
- 1. ASSUMPTIONS & 2. CHECKLIST: Auditable preliminary outputs leading up to the final decision.
🎯 Interview Focus & Questions
Q: Why ask for visible rationale rather than relying on hidden internal model reasoning?
A: Visible rationale can be parsed, logged, audited by human compliance teams, and validated against ground-truth business logic.
20. Structured Reasoning Patterns
🇬🇧 English Definition
Structured reasoning patterns require models to output explicit intermediate artifacts such as criteria tables, rubric checks, or test cases before giving the final answer.
🇮🇳 Hinglish Definition
Structured reasoning mein model se evaluation tables, rubric rubrics, ya test cases pehle generate karwaye jaate hain final output se pehle.
// Structured Reasoning Rubric Prompt
structured_reasoning = """Evaluate the candidate response using this Rubric:
- Technical Accuracy (0-5)
- Code Safety (0-5)
- Clarity (0-5)
Output the Rubric Table first, then provide the Overall Score and Feedback."""
print(structured_reasoning)
🔍 Line-by-Line Prompt / Code Breakdown
- Rubric Table First: Forces quantitative scoring before generating qualitative feedback.
🎯 Interview Focus & Questions
Q: What is the primary benefit of structured reasoning patterns in production?
A: They transform subjective AI generations into structured, quantitative, and reproducible evaluation outputs.
21. Chain-of-Thought (CoT) vs Concise Rationale
🇬🇧 English Definition
Chain-of-Thought (CoT) instructs the model to 'think step by step' through complex logic. Concise Rationale requests a short 1-2 sentence evidence summary for low-latency production APIs.
🇮🇳 Hinglish Definition
CoT model ko 'think step by step' karke long internal reasoning karne bolta hai. Production APIs ke liye Concise Rationale (1-2 lines evidence summary) latency & cost save karta hai.
// CoT vs Concise Rationale Comparison
cot_prompt = "Solve this math word problem. Think step by step and show all intermediate equations before the final number."
concise_rationale_prompt = "Solve this math word problem. Provide a 1-sentence calculation summary followed by the final number."
print("CoT Prompt:", cot_prompt)
print("Concise Rationale Prompt:", concise_rationale_prompt)
🔍 Line-by-Line Prompt / Code Breakdown
- CoT vs Concise: Demonstrates the trade-off between thorough step-by-step reasoning and fast, low-cost API responses.
🎯 Interview Focus & Questions
Q: When is Chain-of-Thought (CoT) essential?
A: For multi-step math word problems, complex logic puzzles, multi-file code debugging, and symbolic reasoning tasks.
22. Prompt Chaining
🇬🇧 English Definition
Prompt Chaining breaks complex workflows into multiple independent API calls where the output of Prompt N becomes the input context for Prompt N+1.
🇮🇳 Hinglish Definition
Prompt Chaining mein ek bada task multiple sequential API calls mein break hota hai (e.g. Prompt 1: Extract facts -> Prompt 2: Write Code -> Prompt 3: Review Code).
// Prompt Chaining Workflow Pseudocode
def prompt_chain_pipeline(raw_doc):
// Call 1: Extract Requirements
reqs = call_llm(f"Extract key requirements from: {raw_doc}")
// Call 2: Generate Solution
solution = call_llm(f"Generate Python code for requirements: {reqs}")
// Call 3: Review Solution
review = call_llm(f"Review this code for security vulnerabilities: {solution}")
return solution, review
print("Prompt Chain Pipeline Configured.")
🔍 Line-by-Line Prompt / Code Breakdown
- prompt_chain_pipeline: Decouples extraction, generation, and code review into 3 modular, testable API steps.
🎯 Interview Focus & Questions
Q: Why is Prompt Chaining superior to writing a single massive prompt for complex workflows?
A: Chaining improves reliability, allows independent error handling/retries per step, keeps context windows focused, and simplifies prompt debugging.
23. Self-Critique & Review Prompts
🇬🇧 English Definition
Self-critique prompts instruct the model to review a draft against explicit quality rubrics and list concrete bug fixes before generating the final revised response.
🇮🇳 Hinglish Definition
Self-critique prompt model ko bolta hai ki wo apne initial draft ko check kare, errors list kare, aur fir clean updated final version deliver kare.
// Self-Critique Prompt Template
critique_prompt = """Review your previous code draft against these criteria:
1. Are all open file handles closed using context managers?
2. Is error handling implemented for HTTP timeout?
List any missing criteria, then output the REVISED_CODE."""
print(critique_prompt)
🔍 Line-by-Line Prompt / Code Breakdown
- REVISED_CODE: Forces self-correction pass before emitting final code block.
🎯 Interview Focus & Questions
Q: What is the Critic-Generator pattern?
A: A design pattern where one prompt (Generator) creates content and a separate prompt (Critic) evaluates and refines it against strict rules.
24. Role + Task + Context Template
🇬🇧 English Definition
The RTC Template (Role, Task, Context, Inputs, Constraints, Output) is the foundational baseline framework for structuring reliable production prompts.
🇮🇳 Hinglish Definition
RTC (Role + Task + Context) baseline template hai jo clear structure, constraints, aur expected output layout guarantee karta hai.
// RTC Baseline Template Construction
rtc_template = """ROLE: Senior DevOps Engineer
TASK: Create a Dockerfile for a FastAPI Python 3.12 application.
CONTEXT: Production Kubernetes cluster running on AWS EKS.
INPUTS: Requirements file contains pandas, numpy, and uvicorn.
CONSTRAINTS: Multi-stage build required to keep image size < 200MB.
OUTPUT: Single Dockerfile block with comments."""
print(rtc_template)
🔍 Line-by-Line Prompt / Code Breakdown
- rtc_template: Practical production framework covering all core prompt requirements cleanly.
🎯 Interview Focus & Questions
Q: Why is a structured template better than free-form paragraph prompting?
A: Structured templates separate concerns clearly, making prompts easier to version-control, audit, and programmatically construct.
25. The CO-STAR Framework
🇬🇧 English Definition
CO-STAR stands for Context, Objective, Style, Tone, Audience, and Response Format — a structured framework for framing non-technical and creative prompts.
🇮🇳 Hinglish Definition
CO-STAR Framework: Context, Objective, Style, Tone, Audience, Response Format. Ye non-technical and communication prompts ke liye popular framework hai.
// CO-STAR Framework Example
costar_prompt = """C - Context: Product launch for an AI-powered fitness app.
O - Objective: Write a 3-email onboarding sequence for new users.
S - Style: Energetic, motivational, concise.
T - Tone: Friendly professional coach.
A - Audience: Busy working professionals aged 25-40.
R - Response Format: Markdown emails with Subject Line, Body, and Call to Action."""
print(costar_prompt)
🔍 Line-by-Line Prompt / Code Breakdown
- CO-STAR Breakdown: Maps communication requirements across Context, Objective, Style, Tone, Audience, and Response format.
🎯 Interview Focus & Questions
Q: What is the main advantage of the CO-STAR framework?
A: It ensures that key creative variables (style, tone, target audience) are explicitly defined rather than left to model guesswork.
26. R-T-C-O-Q Framework
🇬🇧 English Definition
R-T-C-O-Q stands for Role, Task, Context, Output Format, and Quality Criteria — a developer-focused framework for technical prompts.
🇮🇳 Hinglish Definition
R-T-C-O-Q Framework: Role, Task, Context, Output Format, Quality Criteria. Ye technical prompts aur software engineering tasks ke liye best template hai.
// RTCOQ Framework Template
rtcoq_prompt = """R - Role: Senior PostgreSQL Specialist
T - Task: Write an optimized SQL query for top 10 customers by revenue
C - Context: Table 'orders' has 10M rows with indexes on customer_id and created_at
O - Output: EXPLAIN ANALYZE SQL query block
Q - Quality: Zero full-table scans allowed"""
print(rtcoq_prompt)
🔍 Line-by-Line Prompt / Code Breakdown
- RTCOQ: Combines technical role and task with quantitative quality criteria ('Zero full-table scans').
🎯 Interview Focus & Questions
Q: How does adding Quality Criteria (Q) change SQL generation?
A: It forces the model to select indexed join paths and query structures optimized for performance rather than naive SQL syntax.
27. CLEAR Prompt Framework
🇬🇧 English Definition
CLEAR stands for Context, Limits, Examples, Action, and Result — a versatile framework designed to balance constraints with clear action targets.
🇮🇳 Hinglish Definition
CLEAR Framework: Context, Limits (Constraints), Examples, Action, Result (Output). Ye prompt clarity and boundary enforcement maintain karta hai.
// CLEAR Framework Implementation
clear_prompt = """C - Context: Customer support escalation for billing error.
L - Limits: Do not promise refunds over $50 without manager approval.
E - Examples: [Input: Charge $30 -> Output: Instant Refund]
A - Action: Draft customer apology response.
R - Result: Markdown email draft with escalation flag if needed."""
print(clear_prompt)
🔍 Line-by-Line Prompt / Code Breakdown
- CLEAR Framework: Ensures business policy limits and examples directly inform the action target.
🎯 Interview Focus & Questions
Q: What is the role of Limits in the CLEAR framework?
A: Limits define compliance boundaries and policy ceilings, preventing models from making unauthorized business commitments.