03. Domain Prompts: Coding, SQL, Data, RAG & Agents

28. Prompt for Summarization

🇬🇧 English Definition

Summarization prompts must explicitly specify source material, target audience, word/bullet constraints, information to preserve/omit, and strict source-fidelity rules.

🇮🇳 Hinglish Definition

Summarization prompt mein source text, length limit, target audience, preserve karne wali details, aur 'do not invent facts' constraint mention hona chahiye.

// Summarization Prompt Template
summarize_prompt = """Task: Summarize the attached transcript.
Audience: Executive Board Members.
Length: Maximum 5 bullet points, under 150 words total.
Preserve: Key financial figures, dates, and risk decisions.
Omit: Informal small talk and introductory pleasantries.
Constraint: Rely ONLY on the text below. Do not add outside knowledge.


{TRANSCRIPT_TEXT}
"""

print(summarize_prompt)
            

🔍 Line-by-Line Prompt / Code Breakdown

Preserve vs Omit: Gives explicit filtering criteria to select relevant details from long source texts.

🎯 Interview Focus & Questions

Q: How do you prevent a summarization prompt from adding hallucinated background details?
A: Include a strict grounding constraint: 'Rely ONLY on the provided text. Do not assume or extrapolate unstated facts.'

29. Prompt for Extraction

🇬🇧 English Definition

Extraction prompts require explicit field names, data types, missing value rules, and strict schema formatting (such as JSON) to turn unstructured text into clean data.

🇮🇳 Hinglish Definition

Extraction prompt unstructured text se specific fields (Name, Date, Amount) nikalta hai into a clean JSON format. Missing fields ke liye explicit 'null' rule specify karein.

// Structured Data Extraction Prompt
extraction_prompt = """Extract customer details from the email below.

Return ONLY a valid JSON object matching this schema:
{
    "customer_name": "string or null",
    "order_number": "string or null",
    "issue_category": "REFUND | LATE_DELIVERY | WRONG_ITEM | OTHER",
    "urgency_score": "integer 1 to 5"
}

Rule: If a field is absent, set value to null. Do not guess missing numbers.


Hi, my order ID-98765 was delayed by 3 days. Please help. - Amit
"""

print(extraction_prompt)
            

🔍 Line-by-Line Prompt / Code Breakdown

JSON Schema & Null Rule: Explicitly dictates keys, allowed enums, and missing value behavior (`null`).

🎯 Interview Focus & Questions

Q: Why is specifying missing-value behavior essential in extraction prompts?
A: If missing value behavior is omitted, LLMs often fabricate realistic-looking placeholders (e.g. inventing an order ID when none exists).

30. Prompt for Classification

🇬🇧 English Definition

Classification prompts define explicit label names, label descriptions, decision boundaries, edge-case handling rules, and few-shot examples for ambiguous categories.

🇮🇳 Hinglish Definition

Classification prompt categories define karta hai (e.g. SPAM, HAM, INQUIRY) with decision boundaries. Edge-case labeling logic define karna zaroori hai.

// Classification Prompt Template
classification_prompt = """Classify the support ticket into EXACTLY ONE category:
- BILLING: Invoices, payment failures, subscription changes.
- TECHNICAL: App crashes, bug reports, API errors.
- GENERAL: Feature requests, account settings, general questions.

Rule: If a ticket mentions both billing and technical crash, prioritize TECHNICAL.

Ticket: "I cannot update my credit card and the checkout page crashes with error 500."
Category:"""

print(classification_prompt)
            

🔍 Line-by-Line Prompt / Code Breakdown

Prioritization Rule: Resolves multi-category ambiguity by providing explicit tie-breaking logic.

🎯 Interview Focus & Questions

Q: How do you handle edge cases where a text belongs to multiple classification categories?
A: Define explicit priority hierarchy rules (e.g. 'Prioritize TECHNICAL over BILLING when both appear') or allow multi-label classification.

31. Prompt for Rewriting

🇬🇧 English Definition

Rewriting prompts specify original text, target tone, target audience, forbidden terms, length constraints, and whether factual information must remain strictly invariant.

🇮🇳 Hinglish Definition

Rewriting prompt text ka tone, clarity ya format badalta hai (e.g. Technical -> ELI5) while keeping facts strictly unchanged.

// Text Rewriting Prompt Template
rewrite_prompt = """Rewrite the technical release note for a non-technical marketing email.
Target Audience: End consumers.
Tone: Friendly, excited, easy to understand.
Constraint: Keep all feature names and dates 100% accurate. Do not change facts.

Original: "Version 2.4 deploys non-blocking async IO sockets, reducing server P99 latency by 45%."
Rewritten Email:"""

print(rewrite_prompt)
            

🔍 Line-by-Line Prompt / Code Breakdown

Tone & Audience: Transforms technical jargon into consumer-friendly marketing prose while preserving facts.

🎯 Interview Focus & Questions

Q: What is the main risk when asking an LLM to rewrite text?
A: Factual distortion or drift, where the model accidentally alters numbers, dates, or technical facts while trying to improve tone.

32. Prompt for Translation

🇬🇧 English Definition

Translation prompts define source and target languages, regional dialect, formality level, audience context, and rules for preserving technical terms, brand names, and code blocks.

🇮🇳 Hinglish Definition

Translation prompt language, dialect, formality (Formal/Informal) aur brand names/code blocks ko un-translated rakhne ka constraint specify karta hai.

// Technical Translation Prompt
translation_prompt = """Translate the following documentation from English to Hindi.
Formality: Professional / Formal (Aap / Technical Hindi).
Constraint: Preserve code blocks, variables, and brand names in original English.

Source: "Deploy the FastAPI container using Docker on AWS ECS."
Translation:"""

print(translation_prompt)
            

🔍 Line-by-Line Prompt / Code Breakdown

Preserve English Terms: Prevents weird translations of technical terms like 'FastAPI' or 'Docker'.

🎯 Interview Focus & Questions

Q: How do you handle domain-specific terminology during translation?
A: Provide a glossary dictionary in the prompt (e.g. 'Kubernetes -> Kubernetes', 'Pod -> Pod') to prevent inappropriate literal translations.

33. Prompt for Coding

🇬🇧 English Definition

Coding prompts must state programming language, exact version, environment specs, input/output types, algorithmic constraints, code style, and test requirements.

🇮🇳 Hinglish Definition

Coding prompt mein Python version, libraries, input/output data types, constraints (e.g. O(N) time complexity) aur unit test requirement likhein.

// Code Generation Prompt Template
coding_prompt = """ROLE: Senior Python Engineer
TASK: Write a function to check if a binary tree is height-balanced.
ENVIRONMENT: Python 3.12, type hints required.
CONSTRAINTS: Time complexity must be O(N), Space complexity O(H).
OUTPUT: Complete Python code with docstring, type annotations, and 2 unit tests."""

print(coding_prompt)
            

🔍 Line-by-Line Prompt / Code Breakdown

ENVIRONMENT & CONSTRAINTS: Specifies Python 3.12 type hints and O(N) time complexity bounds.

🎯 Interview Focus & Questions

Q: Why is specifying the exact environment and library versions essential for code generation?
A: Language syntax and library methods change across versions (e.g. Pydantic v1 vs v2, Python 3.8 vs 3.12); specifying versions prevents deprecated code generation.

34. Prompt for Debugging

🇬🇧 English Definition

Debugging prompts require providing exact error stack traces, relevant code snippets, runtime environment details, and expected vs actual behavior.

🇮🇳 Hinglish Definition

Debugging prompt mein exact error message, stack trace, original code aur runtime environment dekar root cause + minimal safe fix maangein.

// Code Debugging Prompt Template
debug_prompt = """ROLE: Expert Python Debugger
TASK: Diagnose and fix the error below.
ENVIRONMENT: Python 3.11, FastAPI 0.100.

CODE:

@app.get("/users/{user_id}")
def get_user(user_id: int):
    return await db.fetch_user(user_id)


ERROR:

SyntaxError: 'await' outside async function


REQUIREMENTS:
1. Explain root cause in 2 sentences.
2. Provide corrected code snippet.
3. Do not rewrite unrelated lines."""

print(debug_prompt)
            

🔍 Line-by-Line Prompt / Code Breakdown

CODE & ERROR: Isolated within XML tags to give exact reproducible evidence for debugging.

🎯 Interview Focus & Questions

Q: What constraint prevents LLMs from rewriting your entire codebase during a bug fix?
A: Include the constraint: 'Provide a minimal safe fix. Do not rewrite or modify unrelated working code.'

35. Prompt for SQL

🇬🇧 English Definition

SQL prompts must supply database dialect (PostgreSQL, MySQL, Snowflake), schema definitions, sample table rows, performance constraints, and explicit output formatting rules.

🇮🇳 Hinglish Definition

SQL prompt mein DB dialect (PostgreSQL 15), table schema with indexes, sample data, and performance rules ('Only SELECT, no DELETE') zaroori hain.

// SQL Generation Prompt Template
sql_prompt = """ROLE: Lead Database Architect
DIALECT: PostgreSQL 15

SCHEMA:
CREATE TABLE users (id INT PRIMARY KEY, name VARCHAR(100), created_at TIMESTAMP);
CREATE TABLE orders (id INT PRIMARY KEY, user_id INT, amount DECIMAL(10,2), status VARCHAR(20));

TASK: Write a query to find the top 5 users by total completed order amount in 2026.
CONSTRAINTS: Return ONLY executable SQL. Use INNER JOIN and index-friendly WHERE filtering."""

print(sql_prompt)
            

🔍 Line-by-Line Prompt / Code Breakdown

SCHEMA & CONSTRAINTS: Provides CREATE TABLE DDLs and forces raw SQL generation without conversational text.

🎯 Interview Focus & Questions

Q: How do you prevent SQL injection and unauthorized data modification in SQL prompts?
A: Add strict constraints: 'Generate READ-ONLY SELECT queries. Do not generate INSERT, UPDATE, DELETE, or DROP statements.'

36. Prompt for Data Analysis

🇬🇧 English Definition

Data analysis prompts specify dataset schema, key performance indicators (KPIs), analytical objectives, business context, statistical checks, and required visualization recommendations.

🇮🇳 Hinglish Definition

Data Analysis prompt dataset schema, target business KPI, analysis goal, aur expected deliverable (e.g. Summary, Top 3 findings, Chart recommendations) define karta hai.

// Data Analysis Prompt Template
data_analysis_prompt = """ROLE: Senior Data Analyst
TASK: Analyze customer churn data and identify top 3 churn drivers.
CONTEXT: SaaS platform with monthly recurring subscription model.
INPUT SCHEMA: user_id, tenure_months, support_tickets, monthly_spend, churned (0/1).
OUTPUT:
1. Executive Summary (100 words)
2. Top 3 Data Findings with evidence
3. Recommended retention actions"""

print(data_analysis_prompt)
            

🔍 Line-by-Line Prompt / Code Breakdown

INPUT SCHEMA & OUTPUT: Guides model from raw schema to executive insights and business retention recommendations.

🎯 Interview Focus & Questions

Q: How do you instruct an LLM to avoid drawing false causal conclusions in data analysis?
A: Include the constraint: 'Distinguish correlation from causation. State limitations and underlying data assumptions explicitly.'

37. Prompt for Research

🇬🇧 English Definition

Research prompts define research topic, geographic/temporal scope, source quality rules, mandatory citations, comparison matrices, and explicit uncertainty handling.

🇮🇳 Hinglish Definition

Research prompt research topic, date range, authoritative source guidelines, citation rules, aur unknown/uncertain facts ko explicitly flag karne ka instruction deta hai.

// Technical Research Prompt Template
research_prompt = """TASK: Compare Vector Database indexing algorithms (HNSW vs IVF-PQ).
SCOPE: High-concurrency production deployments (10M+ vectors).
CRITERIA: Build a Markdown comparison table across Recall Rate, Search Latency, Index Build Time, and VRAM Memory Usage.
REQUIREMENTS: Cite authoritative papers/benchmarks. Explicitly state trade-offs."""

print(research_prompt)
            

🔍 Line-by-Line Prompt / Code Breakdown

Markdown comparison table: Forces structured side-by-side evaluation across defined technical criteria.

🎯 Interview Focus & Questions

Q: How do you prevent AI research prompts from citing non-existent papers?
A: Require URL/DOI verification, use Web/RAG search tools, and instruct: 'If specific citation details are unverified, label them as [Citation Needed].'

38. Prompt for Web / Tool Use

🇬🇧 English Definition

Tool use prompts define conditions when external web search or API tools should be triggered, required input schemas, and how to synthesize tool results into final answers.

🇮🇳 Hinglish Definition

Tool Use prompt LLM ko batata hai ki external web search ya tool kab invoke karna hai aur search results ko response mein kaise verify karke cite karna hai.

// Tool Use Instruction Prompt
tool_use_prompt = """AVAILABLE TOOLS:
- web_search(query): Searches live Google web index.
- get_weather(city): Fetches current weather.

POLICY:
- Use web_search ONLY when answering questions about current events after 2024.
- Always cite verified search result URLs in response.

User Question: What were the top tech news headlines today?"""

print(tool_use_prompt)
            

🔍 Line-by-Line Prompt / Code Breakdown

AVAILABLE TOOLS & POLICY: Defines exact tool capabilities and triggers search only for real-time post-2024 questions.

🎯 Interview Focus & Questions

Q: How do you handle tool execution failures gracefully in prompts?
A: Instruct the model: 'If tool execution returns an error or empty result, inform the user and ask for clarification rather than hallucinating answers.'

39. Prompt for RAG

🇬🇧 English Definition

RAG prompts supply retrieved context documents, establish strict source-grounding rules, mandate inline citations, and specify exact fallback behavior when context is insufficient.

🇮🇳 Hinglish Definition

RAG prompt retrieved text chunks context mein supply karta hai and strict rule deta hai: 'Ans only using context. If context doesn't contain answer, say context insufficient.'

// Production RAG Prompt Template
rag_prompt = """SYSTEM POLICY: You are an enterprise Q&A assistant.
Answer factual queries relying ONLY on the provided context documents.

RULES:
1. Cite source document IDs inline like [Doc 1].
2. If context is insufficient, respond: "The provided context does not contain enough information to answer."
3. Do NOT use prior knowledge outside the context.

{RETRIEVED_CHUNKS}

Question: {USER_QUESTION}"""

print(rag_prompt)

🔍 Line-by-Line Prompt / Code Breakdown

RULES 1-3: Establishes strict grounding, inline citations `[Doc 1]`, and standardized fallback text.

🎯 Interview Focus & Questions

Q: What is the single most important rule in a RAG prompt?
A: The strict fallback rule ('If context is insufficient, state so') which prevents the model from falling back on generic pre-trained knowledge.

40. Prompt for RAG Grounding

🇬🇧 English Definition

RAG Grounding prompts explicitly distinguish between direct facts stated in context and model-derived inferences, preventing unsupported claims.

🇮🇳 Hinglish Definition

RAG Grounding prompt direct facts aur assumptions ke beech clear separation force karta hai taaki corporate compliance requirements complete ho sakein.

// Grounding Differentiation Prompt
grounding_prompt = """Format your response into two distinct sections:
1. DIRECT EVIDENCE: Exact facts supported directly by the context chunks (with citations).
2. INFERRED LOGIC: Any logical deductions or assumptions derived from the evidence.

If no direct evidence exists, list 'NONE' under Direct Evidence."""

print(grounding_prompt)
            

🔍 Line-by-Line Prompt / Code Breakdown

DIRECT EVIDENCE vs INFERRED LOGIC: Forces clear segregation of backed facts from probabilistic reasoning.

🎯 Interview Focus & Questions

Q: How does Grounding Evaluation differ from Retrieval Evaluation?
A: Retrieval evaluation checks if the right documents were found; Grounding evaluation checks if the final LLM response is strictly truthful to those retrieved documents.

41. Prompt for Agents

🇬🇧 English Definition

Agent prompts define identity, high-level goal, available tool schemas, step-by-step decision rules, tool permission boundaries, and explicit stop conditions.

🇮🇳 Hinglish Definition

Agent prompt Agent identity, high-level goal, tools list, decision rules, permission boundaries aur explicit stop condition define karta hai.

// Autonomous Agent System Prompt
agent_prompt = """IDENTITY: Autonomous Customer Refund Agent
GOAL: Investigate customer refund request and issue refund if criteria met.

AVAILABLE TOOLS:
- check_order_status(order_id)
- verify_payment(transaction_id)
- issue_refund(order_id, amount)

RULES:
- Never issue refund exceeding $100 without manager approval.
- Stop execution after calling issue_refund and output final summary.

STOP CONDITION: Refund issued OR request escalated to human manager."""

print(agent_prompt)
            

🔍 Line-by-Line Prompt / Code Breakdown

STOP CONDITION & RULES: Prevents infinite agent loops and enforces financial spending ceilings ($100).

🎯 Interview Focus & Questions

Q: Why do AI Agents require explicit stop conditions in their system prompts?
A: Without explicit stop conditions, autonomous agent loops can get trapped in repetitive tool execution loops, draining API budgets.

42. Tool Calling

🇬🇧 English Definition

Tool calling prompts give tools unambiguous names, descriptive descriptions, explicit parameter type hints, and usage constraints for accurate model function selection.

🇮🇳 Hinglish Definition

Tool Calling prompts tools ko clear names, detailed descriptions aur parameter types dete hain taaki model sahi tool pick kare.

// Tool Definition Protocol
tool_def = {
    "name": "calculate_mortgage",
    "description": "Calculates monthly mortgage payments based on principal, annual interest rate, and term years.",
    "parameters": {
        "principal": {"type": "float", "description": "Loan amount in USD"},
        "rate": {"type": "float", "description": "Annual interest rate percentage e.g. 6.5"},
        "years": {"type": "int", "description": "Loan duration in years e.g. 30"}
    },
    "required": ["principal", "rate", "years"]
}

print("Tool Definition:", tool_def["name"])
            

🔍 Line-by-Line Prompt / Code Breakdown

tool_def: Clear parameter descriptions and required fields array ensuring the LLM supplies all necessary inputs.

🎯 Interview Focus & Questions

Q: How do descriptive tool descriptions improve function calling accuracy?
A: LLMs select tools by comparing user query semantics against tool description vectors; detailed descriptions prevent wrong tool invocation.

43. Function Calling Schema

🇬🇧 English Definition

Function calling schemas strictly define types, enums, required fields, and parameter boundaries, requiring client-side validation before executing functions.

🇮🇳 Hinglish Definition

Function Calling Schema strict data types & required fields enforce karta hai. App backend ko execution se pehle argument validation zaroor karni chahiye.

// Function Calling Validation Pseudocode
def validate_and_execute_tool(model_call_json):
    tool_name = model_call_json.get("name")
    args = model_call_json.get("arguments")
    
    // Client-side schema validation
    if tool_name == "calculate_mortgage":
        if args["principal"] <= 0 or args["rate"] <= 0:
            return "Error: Arguments must be positive numbers."
        return f"Executing {tool_name} with principal=${args['principal']}"
    return "Unknown Tool"

print(validate_and_execute_tool({"name": "calculate_mortgage", "arguments": {"principal": 300000, "rate": 6.5, "years": 30}}))
            

🔍 Line-by-Line Prompt / Code Breakdown

validate_and_execute_tool: Client-side argument validation protecting application backend from malformed LLM tool arguments.

🎯 Interview Focus & Questions

Q: Why should you never execute LLM function calls without client-side validation?
A: Because LLMs can hallucinate invalid arguments, out-of-range numbers, or wrong parameter types that crash backend services.

44. JSON / Structured Output

🇬🇧 English Definition

Structured output prompts enforce rigid JSON schemas, defining field keys, data types, allowed enums, and explicit missing-field behavior for reliable backend parsing.

🇮🇳 Hinglish Definition

Structured JSON Prompt model ko exact schema layout mein answer dene bolta hai without extra text, enabling seamless Pydantic parsing in APIs.

// JSON Schema Enforcer Prompt
json_enforcer = """Respond ONLY with a raw JSON object matching this schema:
{
    "status": "SUCCESS | ERROR",
    "data": {
        "items_count": integer,
        "total_cost": float
    },
    "errors": ["string"]
}

Rule: Do not include markdown code block formatting like ```json. Return raw JSON string only."""

print(json_enforcer)
            

🔍 Line-by-Line Prompt / Code Breakdown

Raw JSON Rule: Instructs model to omit markdown backticks for immediate string deserialization.

🎯 Interview Focus & Questions

Q: How do JSON mode and Grammar-Guided Decoding improve output reliability?
A: They constrain token generation probabilities at the API inference layer, ensuring 100% syntactically valid JSON output.