08. Line-by-Line Prompts, 10 Projects & 90-Day Roadmap
Sections 96-107: Line-by-Line Breakdown of Master, Coding, RAG, JSON & Agent Prompts; Evaluation Rubrics; Debugging Examples; 10 Projects; 20 Interview Q&A; 90-Day Roadmap; Audit Checklist & The Golden Rule.
96. Master Prompt Template — Line by Line
🇬🇧 English Definition
A line-by-line breakdown of a production Master Prompt Template spanning Role, Objective, Context, Input Delimiters, Constraints, and Structured Output.
🇮🇳 Hinglish Definition
Line-by-Line Master Prompt Explanation: Role (Perspective), Objective (Exact Task), Context (Business Goal), Input (Delimited Data), Constraints (Boundaries), Output (Exact Structure).
ROLE: You are a senior data analyst.
OBJECTIVE: Analyze the provided sales data and identify the three biggest drivers of declining revenue.
CONTEXT: The business sells products online. The audience is the management team.
INPUT:
{{DATA}}
CONSTRAINTS:
- Use only the provided data.
- Do not invent missing values.
- State assumptions explicitly.
- Keep the response under 500 words.
OUTPUT:
1. Executive summary
2. Top 3 findings
3. Evidence for each finding
4. Recommended actions
5. Data limitations
🔍 Line-by-Line Prompt / Code Breakdown
- ROLE: Sets senior data analyst perspective.
- OBJECTIVE: Identifies top 3 declining revenue drivers.
- CONTEXT: Online sales business context for executive audience.
- INPUT: Delimits raw sales data safely inside `
`. - CONSTRAINTS: Bans inventing missing values and caps response under 500 words.
- OUTPUT: Dictates exact 5-part structure for easy evaluation.
🎯 Interview Focus & Questions
Q: Why does line-by-line breakdown of templates improve engineering rigor?
A: Because every single line in a production prompt serves an explicit functional purpose: establishing role, bounding context, preventing hallucinations, or dictating output schema.
97. Coding Prompt — Line by Line
🇬🇧 English Definition
A line-by-line breakdown of a production Code Debugging & Fixing Prompt covering Role, Task, Environment, Code, Error Stack, and Minimal-Fix Requirements.
🇮🇳 Hinglish Definition
Line-by-Line Coding Prompt: Role (Senior Dev), Task (Fix Bug), Environment (Python 3.12, FastAPI), Code/Error (Delimited evidence), Requirements (Minimal fix + Unit test).
ROLE: Act as a senior Python developer.
TASK: Fix the bug in the code below.
ENVIRONMENT: Python 3.12, FastAPI, Windows.
CODE:
```python
{{CODE}}
```
ERROR:
```text
{{ERROR}}
```
REQUIREMENTS:
- Identify the root cause.
- Provide the minimal safe fix.
- Return corrected code.
- Explain each changed line.
- Add one test case.
- Do not rewrite unrelated code.
🔍 Line-by-Line Prompt / Code Breakdown
- ENVIRONMENT: Specifies Python 3.12, FastAPI, Windows runtime.
- CODE / ERROR: Encloses code and error log inside code fences.
- REQUIREMENTS: Demands root cause, minimal fix, line explanations, test case, and bans rewriting unrelated code.
🎯 Interview Focus & Questions
Q: Why is 'Do not rewrite unrelated code' a critical line in coding prompts?
A: Without this line, LLMs frequently refactor surrounding variable names, reformat style, or break working helper functions.
98. RAG Prompt — Grounded Answering
🇬🇧 English Definition
A line-by-line breakdown of a Production RAG Grounding Prompt establishing system policy, delimited context, user question, and strict fallback rules.
🇮🇳 Hinglish Definition
Line-by-Line RAG Grounded Prompt: System Policy (Answer only using context), Context (`
SYSTEM POLICY: Answer factual questions using the supplied context.
CONTEXT:
{{RETRIEVED_DOCUMENTS}}
USER QUESTION:
{{QUESTION}}
RULES:
1. Use the context for factual claims.
2. If the context is insufficient, say: "The provided context is insufficient."
3. Do not invent citations.
4. Keep the answer concise.
5. Distinguish direct evidence from inference.
🔍 Line-by-Line Prompt / Code Breakdown
- SYSTEM POLICY: Dictates grounded behavior.
- CONTEXT: Delimits retrieved chunks.
- RULES 1-5: Standardizes exact fallback sentence, bans fake citations, and requires evidence separation.
🎯 Interview Focus & Questions
Q: What makes this RAG prompt robust against hallucinations?
A: The combination of XML context isolation, strict non-extrapolation rules, and a mandatory exact fallback phrase when context lacks evidence.
99. Structured JSON Prompt
🇬🇧 English Definition
A line-by-line breakdown of a Structured Extraction Prompt enforcing raw JSON output schemas, null field handling, and enum value constraints.
🇮🇳 Hinglish Definition
Line-by-Line Structured JSON Prompt: Task (Extract info), Schema (Exact JSON shape), Rules (Do not guess, Use null for missing, Enum validation).
TASK: Extract customer information from the text.
RETURN ONLY JSON:
{
"name": "string or null",
"email": "string or null",
"phone": "string or null",
"intent": "sales|support|billing|other"
}
RULES:
- Do not guess.
- Use null when a field is absent.
- intent must be one of the listed values.
- Preserve the customer's spelling.
🔍 Line-by-Line Prompt / Code Breakdown
- RETURN ONLY JSON: Forces pure JSON response layout.
- Schema: Specifies allowed data types and enum options for `intent`.
- RULES: Anti-guessing, null handling, and exact spelling preservation rules.
🎯 Interview Focus & Questions
Q: Why specify 'intent must be one of the listed values'?
A: Enforcing exact enum choices prevents the model from generating custom intent labels (e.g. 'refund_inquiry') that break downstream backend switch statements.
100. Agent Prompt
🇬🇧 English Definition
A line-by-line breakdown of an Autonomous Agent System Prompt defining Role, Goal, Tools List, Execution Rules, and Stop Conditions.
🇮🇳 Hinglish Definition
Line-by-Line Agent Prompt: Role (Support assistant), Goal (Resolve issue), Tools (search_kb, create_ticket), Rules (Search before policy answer, Never invent policy), Stop Condition.
ROLE: You are a customer-support assistant.
GOAL: Resolve the user's issue using approved tools.
TOOLS:
- search_kb(query): search the knowledge base.
- create_ticket(summary): create a support ticket.
RULES:
- Search the knowledge base before answering policy questions.
- Never invent company policy.
- Ask for missing required information.
- Create a ticket only when escalation criteria are met.
- Do not expose internal tool details.
STOP CONDITION: Return a concise customer-facing answer after the issue is resolved or escalation is created.
🔍 Line-by-Line Prompt / Code Breakdown
- ROLE & GOAL: Establishes identity and task target.
- TOOLS: Explicitly lists tool names and signatures.
- RULES: Mandatory search policy and anti-invention constraint.
- STOP CONDITION: Prevents infinite tool loops by defining exact task completion state.
🎯 Interview Focus & Questions
Q: Why is 'Do not expose internal tool details' included in agent prompts?
A: To prevent technical leakages (e.g. exposing internal function names or database keys) in end-user chat interfaces.
101. Prompt Evaluation Rubric
🇬🇧 English Definition
A complete 0-to-2 quantitative scoring rubric evaluating outputs across Accuracy, Completeness, Format, and Grounding.
🇮🇳 Hinglish Definition
Prompt Evaluation Rubric: Accuracy (0-2), Completeness (0-2), Format (0-2), Grounding (0-2). Total = 8 Points.
Score each response from 0 to 2:
Accuracy:
0 = incorrect | 1 = partly correct | 2 = correct
Completeness:
0 = major requirements missing | 1 = some requirements missing | 2 = all required elements present
Format:
0 = unusable format | 1 = minor deviations | 2 = exact required format
Grounding:
0 = unsupported claims | 1 = mixed support | 2 = claims supported by supplied evidence
Total = Accuracy + Completeness + Format + Grounding (Max Score = 8)
🔍 Line-by-Line Prompt / Code Breakdown
- Rubric Matrix: Evaluates 4 dimensions on a 0-2 scale, yielding a composite score out of 8.
🎯 Interview Focus & Questions
Q: How does a quantitative rubric help in prompt engineering?
A: It turns subjective impressions ('the output looks okay') into measurable data points to compare candidate prompts against baseline prompts.
102. Prompt Debugging Example
🇬🇧 English Definition
A concrete before-and-after debugging example comparing a weak unconstrained prompt with an improved, schema-enforced prompt.
🇮🇳 Hinglish Definition
Prompt Debugging Example: Problem (Model invents missing details). Weak Prompt ('Extract customer info') vs Improved Prompt ('Extract name, email, phone. Return null if absent. Return ONLY JSON').
Problem: The model keeps inventing missing customer details.
Weak prompt:
"Extract customer information."
Improved prompt:
"Extract name, email and phone from the supplied text.
Return null when a field is absent.
Never infer or invent values.
Return only the defined JSON schema."
Why weak prompt fails: missing schema and missing-value behavior.
Why improved prompt is stronger: fields, fallback, anti-guessing rule and output contract are explicit.
🔍 Line-by-Line Prompt / Code Breakdown
- Weak vs Improved: Highlights how adding schema, null fallback, and anti-guessing rules eliminates hallucinations.
🎯 Interview Focus & Questions
Q: What are the 3 most common reasons a prompt fails in production?
A: 1. Missing explicit schema, 2. Unstated missing-value behavior, 3. Lack of input delimiters separating instructions from user data.
103. Prompt Engineering Projects
🇬🇧 English Definition
10 Industry-standard project blueprints covering Resume Analyzer, PDF Q&A RAG, SQL Assistant, Customer Support Copilot, AI Research Assistant, Code Review Assistant, Meeting Assistant, Data Analyst Agent, Multimodal Document Extractor, and Production Prompt Evaluation Platform.
🇮🇳 Hinglish Definition
Top 10 Prompt Projects: 1. Resume Analyzer, 2. PDF Q&A RAG, 3. SQL Assistant, 4. Customer Support Copilot, 5. AI Research Assistant, 6. Code Review Assistant, 7. Meeting Assistant, 8. Data Analyst Agent, 9. Multimodal Document Extractor, 10. Production Evaluation Platform.
PROJECT BLUEPRINTS SUMMARY:
1. Resume Analyzer: extraction + scoring rubric + JSON output
2. PDF Q&A RAG: retrieval + citations + fallback
3. SQL Assistant: schema-aware SQL generation + validation
4. Customer Support Copilot: RAG + tool calling + escalation
5. AI Research Assistant: search + source evaluation + citations
6. Code Review Assistant: static checks + rubric + test generation
7. Meeting Assistant: transcript extraction + action items
8. Data Analyst Agent: SQL + Python tools + chart recommendations
9. Multimodal Document Extractor: image/PDF + structured JSON
10. Production Evaluation Platform: golden dataset + A/B testing + metrics
🔍 Line-by-Line Prompt / Code Breakdown
- Projects 1-10: Comprehensive practical project portfolio covering all modern LLM architectural patterns.
🎯 Interview Focus & Questions
Q: Which project best demonstrates production prompt engineering skills?
A: Project 10 (Production Evaluation Platform) because it proves you can measure, benchmark, version, and scientifically optimize prompts.
104. Prompt Engineering Interview Questions
🇬🇧 English Definition
20 Core interview questions covering definitions, prompt components, zero-shot vs few-shot, prompt chaining, hallucinations, prompt injection, structured outputs, evaluation, and security.
🇮🇳 Hinglish Definition
Top 20 Technical Interview Questions covering prompt anatomy, few-shot, chaining, injection defense, RAG grounding, evaluation rubrics, and production safety.
TOP INTERVIEW QUESTIONS SAMPLE:
1. What is prompt engineering? -> Systematic practice of designing instructions/context for reliable LLM outputs.
2. Zero-shot vs few-shot? -> Zero-shot has no examples; few-shot provides 2-5 gold examples.
3. How do you reduce hallucinations? -> RAG context grounding, explicit fallback rules, low temperature (0.0).
4. What is prompt injection? -> Untrusted user input attempting to override system instructions.
5. How do you evaluate prompts? -> Run against golden datasets using quantitative rubrics (Accuracy, Format, Grounding).
🔍 Line-by-Line Prompt / Code Breakdown
- Top Questions: High-frequency technical interview questions and concise gold answers.
🎯 Interview Focus & Questions
Q: How do you protect a RAG system from untrusted instructions in ingested documents?
A: Treat retrieved text strictly as passive data inside XML delimiters, harden the system prompt to ignore commands inside data tags, and sanitize inputs.
105. 90-Day Prompt Engineering Roadmap
🇬🇧 English Definition
Structured 90-day learning curriculum: Days 1-10 (LLM basics & tokens), Days 11-20 (Zero/few-shot & roles), Days 21-30 (Coding, SQL & Extraction), Days 31-40 (RAG & Injection), Days 41-50 (Tool calling & Agents), Days 51-60 (Multimodal), Days 61-70 (Evals & A/B testing), Days 71-80 (Versioning & Cost), Days 81-90 (Capstone Application Deployment).
🇮🇳 Hinglish Definition
Day-by-Day 90-Day Study Plan: LLM Basics → Roles & Constraints → Coding & SQL → RAG & Injection → Tools & Agents → Multimodal → Evals & Rubrics → Versioning & Security → Production Capstone App.
90-DAY ROADMAP PHASES:
- Days 1-10: LLM basics, tokens, context, prompt anatomy and output control.
- Days 11-20: zero-shot, few-shot, roles, constraints, examples and structured outputs.
- Days 21-30: coding, SQL, data analysis, summarization and extraction prompts.
- Days 31-40: RAG, grounding, citations and prompt injection.
- Days 41-50: tool calling, function schemas and agents.
- Days 51-60: multimodal prompting and domain-specific workflows.
- Days 61-70: evaluation datasets, rubrics, A/B testing and regression tests.
- Days 71-80: prompt versioning, security, privacy, cost and latency.
- Days 81-90: build a production-style AI application with prompts, tools, evaluation and monitoring.
🔍 Line-by-Line Prompt / Code Breakdown
- Days 1-90: Complete step-by-step learning progression from beginner to production engineer.
🎯 Interview Focus & Questions
Q: What order should you follow when learning Prompt Engineering?
A: Anatomy & Roles -> Constraints & Formats -> Few-Shot & CoT -> Domain Prompts (Coding/SQL) -> RAG Grounding -> Injection Defense -> Evals & Rubrics -> Production LLMOps.
106. Final Prompt Engineering Checklist
🇬🇧 English Definition
A 19-point comprehensive prompt audit checklist covering task specificity, context delimiters, constraint measurability, missing value handling, edge case testing, tool permissions, golden datasets, versioning, fallback, and cost measurement.
🇮🇳 Hinglish Definition
Final 19-Point Audit Checklist: Task specific? Context included? Data isolated? Constraints measurable? Output explicit? Missing values handled? Examples provided? Edge cases tested? Tool permissions enforced? Prompts versioned?
FINAL AUDIT CHECKLIST:
[x] Is the task specific?
[x] Is the required context included?
[x] Are instructions separated from untrusted data?
[x] Are constraints measurable?
[x] Is the output format explicit?
[x] Are missing values handled?
[x] Are edge cases tested?
[x] Is hallucination/grounding behavior defined?
[x] Are tool permissions enforced outside the prompt?
[x] Is there a golden evaluation set & quality rubric?
[x] Is the prompt versioned & monitored in production?
🔍 Line-by-Line Prompt / Code Breakdown
- Audit Checklist: Ensures every production prompt is audited for security, schema, and performance before release.
🎯 Interview Focus & Questions
Q: Why is an audit checklist essential before pushing a prompt to production?
A: Because missing a single item (like data isolation or missing-value rules) can expose the app to prompt injection or parsing crashes.
107. Golden Rule of Prompt Engineering
🇬🇧 English Definition
The Golden Rule: Clear Task + Relevant Context + Explicit Constraints + Examples When Useful + Structured Output + Verification + Evaluation = Better Prompt Engineering.
🇮🇳 Hinglish Definition
Golden Rule: Prompt ko unnecessarily bada nahi, clear aur testable banao. AI ko guess karne ki jagah requirements, context aur expected output provide karo. Production mein prompt ko security boundary mat samjho — application-level validation aur permissions zaroor rakho.
GOLDEN FORMULA:
Clear Task + Relevant Context + Explicit Constraints + Examples When Useful + Structured Output + Verification + Evaluation = Better Prompt Engineering.
Hinglish Summary:
Prompt ko clear aur testable banao. AI ko guess karne ki jagah requirements, context aur expected output provide karo. Production mein prompt ko security boundary mat samjho — application-level validation aur permissions zaroor rakho.
🔍 Line-by-Line Prompt / Code Breakdown
- GOLDEN FORMULA: The ultimate core principle summarizing all 107 sections of Prompt Engineering.
🎯 Interview Focus & Questions
Q: Summarize Prompt Engineering in one sentence for an executive.
A: Prompt engineering is defining clear tasks, context, constraints, and schemas so AI models deliver reliable, secure, and testable business outputs.