🚀 Daily Dose of AI | AI Interview Question of the Day! 🧠 The problem might not be the LLM. It could be OCR (Optical Character Recognition)! Real-World Example: An insurance company uploads 20,000 policy PDFs into its RAG system. Original document: ₹5,00,000 OCR extracts: R5,0O,OOO The AI stores the incorrectly extracted text and later generates an…
🚀 Daily Dose of AI | AI Interview Question of the Day! 🧠 The Answer: RERANKING! Imagine searching for: "Side effects of Drug X in elderly patients." Vector Search: May rank a document about…
🚀 Daily Dose of AI | AI Interview Question of the Day! 🧠 Because Vector Search understands meaning, not necessarily exact matches! Real-World Example (QA): You search your knowledge base for:…
🚀 Daily Dose of AI | What Is AI Agent Observability? 🔍 Your AI Agent completed 10 tasks. But do you know HOW it completed them? That's where AI Agent Observability comes in! Real-World Example…
🚀 Daily Dose of AI | How Does an AI Agent Use Multiple Tools? One instruction. Multiple tools. One final result! Real-World Example (QA): You ask an AI Agent: "Fetch Jira story QA-101, generate test…
Your AI chatbot worked perfectly yesterday. But after a prompt update, it starts giving wrong answers! ❌ PROBLEM: Traditional testing checks fixed outputs. AI can generate different answers for the…
Your employees might already be using AI tools your company doesn't know about! That's called Shadow AI! Real-World Example: A software tester copies a confidential banking requirement document into…
This Diwali, let's light up our knowledge with the ABCs of Artificial Intelligence on Daily Dose of AI! 🚀 🪔 H — Hallucination: When AI confidently generates incorrect information. ✨ A — AI Agents:…
One AI request doesn't always mean one AI call! Real-World Example: You ask an AI Agent: "Analyze this Jira story and generate Playwright tests." 🤖 Behind the scenes, the agent may: Fetch Jira…
Your AI Agent generated 20 test cases. But did it cover the right scenarios? That's where AI Agent Evals come in! Real-World QA Example: Imagine an AI agent generating Playwright tests from a Jira…
Your AI Agent is smart. But what stops it from making a costly mistake? Meet the AI Agent Harness! Think of it this way: AI Model = The Engine Agent Harness = The Safety Belt + Steering + Brakes…
AI can write code in minutes. But can it ensure the code actually works? Vibe Coding — Describe It, Build It! "Create a banking login page." AI generates the code through natural-language…
RAG vs Agentic RAG 🤖 RAG says, “Here's what I found.” Agentic RAG says, “Here's what I figured out for you!” Let's understand this with a simple example: Imagine you're applying for a home loan. You…
Daily Dose of AI | Context Engineering AI isn't always confused. Sometimes you invite it to a meeting without sharing the agenda! Prompt Engineering vs. Context Engineering Prompt Engineering = How…
Daily Dose Of AI | How Does an AI Agent Decide Which Tool to Call? Let's understand with a real-world example! 🧮 Calculator → Solves maths 🌦️ Weather App → Checks weather 🔎 Google Search → Finds…
Daily Dose Of AI | AI Interview Question of the Day! How would you handle an LLM context-window limit? Imagine you have a small plate that can hold only 3 pizza slices, but there are 20 slices…
Skills in Claude Code turn tasks you repeat into reusable workflows. You write the steps once in a SKILL.md file, and from then on you invoke the skill instead of explaining the same instructions…
🚀 Daily Dose of AI | Claude Code What if AI says “Task Completed” without actually testing your code? (Mostly this happens ) Problem: AI-generated code may contain bugs even when Claude says it's…
Daily Dose of AI | Claude Code What if AI could review your code using multiple specialists at once? Problem: A single AI review might miss security, performance, or test-coverage issues. Solution:…
AI tools often say "Done" when the code only looks correct. The fix is to make passing tests the definition of done: Claude writes the tests first, you watch them fail, and then it implements the…
/plan switches Claude Code into Plan Mode. In this mode it explores your project and proposes an implementation plan, and it doesn't change any code until you approve the approach. Daily Dose of AI |…
CLAUDE.md is a project instruction file that Claude Code loads at the start of each session. You write your project's rules once, and you don't have to repeat them in every prompt. Learn the concept…
With Claude Code, Atlassian MCP and Playwright MCP, a single custom command (/story-to-tests QA-101) can fetch a user story, design test cases, generate Page Object Model scripts, run them and…
Claude Code gives you two ways to control what an AI agent can access and run. Permission Deny Rules block specific tools or file paths outright. Hooks run your own scripts before or after an action,…
Copilot can shrink a function from 50 lines to 12, and the tests still pass, but it may remove a business validation that the tests never covered. AI understands your code, but it may not understand…
GitHub Copilot can do more than suggest code. In VS Code it can generate API tests, mock test data, technical documentation, SQL queries and CI workflows, and each of these starts with one…
Five AI tools cover the main QA workflow: ChatGPT for test design and requirements analysis, GitHub Copilot for test automation, Claude Code for MCP-connected use cases, Jira + Rovo for requirements…
RLHF trains AI using human preferences as well as data. People compare answers and choose the more useful one. Those choices train a reward model, and the reward model then steers the AI towards more…
Human-in-the-Loop (HITL) means people review important AI decisions, give feedback and approve critical actions before they happen. AI generates, humans validate, and quality improves. Daily Dose of…
A read-only reviewer agent in GitHub Copilot can search and inspect your code, but its file-editing tools are restricted. It reports bugs, missing validations and missing tests without changing your…
MCP (Model Context Protocol) is a standard way for AI agents to connect to external tools and data sources. With MCP, GitHub Copilot in VS Code can fetch Jira stories, generate and run Playwright…
Next Edit Suggestions (NES) in GitHub Copilot predict where you need to edit next and what the change should be. Inline suggestions only predict what you'll type at the cursor. Daily Dose of AI |…
GitHub Copilot has three customisations. Custom Instructions define how Copilot writes code, Prompt Files define what task it performs repeatedly, and Custom Agents define who performs the task and…
AI guardrails are rules and checks around an AI system that keep it safe, on-topic and predictable. They block sensitive data, limit what the AI answers, enforce output formats and send high-risk…
LLM-as-Judge is a testing method where one AI model scores another model's responses against criteria like accuracy, relevance or tone. It scales evaluation, but the judge can be biased or…
Extracting the text first is usually cheaper, because you send only text tokens. Sending the PDF directly usually costs more, since models like Claude, ChatGPT and Gemini also process each page as an…
To test an AI-generated summary, check three things: faithfulness (no facts that aren't in the source), coverage (no important points missed) and concision (short and clear enough to be useful).…
A golden dataset in GenAI testing is a curated set of real prompts paired with reference answers or expected qualities, used to check AI responses every time the model, prompt or data changes.…
A RAG pipeline works in two stages. First, documents are split into chunks, turned into embeddings and stored in a vector database. Then, when a user asks a question, the system finds the most…
Testing a RAG chatbot? Don’t test it as one big black box. Break it into 3 layers: 🔍 Retrieval → Did it find the right chunks? 🧠 Generation → Is the answer faithful to those chunks? 🔄 End-to-End →…
AI Testing Interview Question You Should Know What is Prompt Injection? It’s when hidden or direct instructions try to make an AI ignore its intended rules and behave differently. QA takeaway: Don’t…
Here’s an important AI Testing interview question: Why should a tester understand the system prompt? Because you can’t test AI effectively without knowing the hidden rules shaping its responses. Know…
👀 AI can now see. 👂 It can hear. 📄 It can read. 🎥 It can even understand video. That’s Multimodal AI—one model connecting meaning across different types of input. Instead of typing a confusing…
A faithful AI answer can still be a BAD answer. In GenAI testing, accuracy alone isn’t enough. A production-ready AI response needs 4 things: 🎯 Faithfulness — Is it grounded? 💡 Relevance — Did it…
AI that doesn’t just answer… but ACTS. We’re moving from: Regular AI: “Here’s your Goa trip itinerary.” ➡️ to Agentic AI: “I found the flights, checked your calendar, booked the hotel, and sent the…
RAGAS vs DeepEval: What’s the Difference? Building an AI/RAG application is one thing. Knowing whether it actually works well is another. Two useful tools for AI evaluation are RAGAS and DeepEval:…
🤖 Your Prompt Is Your Test Case! Traditional testing asks: “What should the user click?” AI testing asks: “What happens when I change the input?” For AI-powered applications, your test suite should…
🔌 How does AI actually connect to GitHub, Slack, Google Drive or other tools? Meet MCP — Model Context Protocol. Think of MCP as USB-C for AI: 🔹 One standard connection 🔹 Multiple tools & data…
Traditional QA vs. GenAI Testing Traditional QA often checks for one expected output. GenAI testing is different—you may have many valid answers. So instead of asking: “Does the response exactly…
What happens when you ask an AI chatbot: “What is your company’s return policy?” Would you trust the answer blindly? Here's the interesting part 👇 ❌ Without RAG AI may answer based on what it…