AI Evals are a hot topic. Any system that uses an LLM is non-deterministic. It can hallucinate or give unsafe responses. The basic solution is using guardrails in the prompt. But they aren't enough. What about unit tests? Useful, but they can't measure subjective metrics like answer relevancy or groundedness. So what do we do? Build an eval-suite. It's a series of tests for output quality and system performance. Because building an agent is easy, but making ROI is the challenge. If you want to know more about how Evals are being used in AI systems, sign up for our cohort at aiengg.dev Cheers!

AI Engineering Cohort registrations are open. The cohort is an 8-week program for professionals who want to build production-grade AI applications. The cohort will run from 6th June to 26th July 2026. ------------ Main topics: Week 1. Basics of LLMs: Tokenization, Vectorization, Attention Week 2. RAG components & architecture Week 3. Advanced RAG & Evals Week 4. RAG Architectures & Specialized Types Week 5. Agentic Systems, ReAct & tool calling Week 6. Multi-Agent Systems, Orchestration Week 7. Context Engineering, Memory & Evals Week 8. Capstone Project: Build your own AI Project ---------------- Primary Audience: Software / AI Engineers Number of Sessions: 16 Live Classes + 7 Networking Sessions Involves Coding: Yes Classes: Live Instructor: Tanishq Singh Pricing: ₹1,00,000 / $1100 ---------------- This cohort is for working professionals looking to gain job-relevant AI capabilities. You can find the detailed syllabus on the website. Register here: aiengg.dev Wishing you a great 2026!

AI Engineering Cohort 2026 Duration: 16 Weeks (28 Feb - 14 Jun) Primary Audience: Software / AI Engineers Registration Link: aiengg.dev

AI research paper #3: Vision Transformers The paper has ~85,000 citations. It explains how transformers classify images at scale, outperforming CNNs and ResNets. Link: youtube.com/live/Tx4HXZ7djlA Why do I read these papers live? Because I think every software engineer should know what's inside them. It helps you build a foundational understanding of AI models. And once you understand how these models work, you can separate reality from hype. For the next three weeks, I will read a research paper live every Wednesday at 6 PM IST. Any suggestions on the topics? Looking forward to the next session.

12 days to launch the AI Engineering Cohort Website. Registrations will start on 31st January 2026. The cohort will start on 28th February 2026. Main topics: Week 0. Setup + Core Math Week 1. Terminology + MNIST Week 2. Basics of LLMs: Tokenization, Vectorization, Attention Week 3. Deep dive into LLMs: QKV matrices, Cross + Self + MH Attention Week 4. LLM Coding: Causal Masking + Code GPT Week 5. Think like an engineer: How massive models are trained to production Week 6. Optimization Hacks: KV Caching, Quantization, LoRA Week 7. The RAG Problem: What is RAG, Chunking, Reranking, Vector DBs Week 8. The RAG Code: Safety + Guardrails, Code RAG Week 9. AI Agents: ReAct Pattern, Tool Calling, LangChain, LangGraph Week 10. Context Engineering: Memory Systems, MCP, Multi-Agents Week 11. AI Engineering: Evals, Tradeoffs, Fine-tuning vs. RAG vs Prompting Week 12. Thinking Models: Reasoning, Chain of Thought Week 13. Multi-modal Models: Images + Video, CLIP, Diffusion Models Week 14. Capstone Project: Build your own AI Project Week 15. Career Goals: How to move to AI Engineering ---------------- Primary Audience: Software / AI Engineers Number of Sessions: 32 Live Classes + 15 General Sessions Involves Coding: Yes Classes: Live Tentative pricing: ₹1,20,000 / $1400 ---------------- This cohort is for working professionals looking to gain job-relevant AI capabilities. The official website will be ready soon. We are working out the final details now. We wish you an amazing year ahead. Cheers!

Mark Zuckerberg: "AI will replace SDE-2s in 2025". Happy New Year. AI still cannot solve complex problems. Software engineers do the heavy lifting. Companies have finally moved away from AI-hype to AI-usage. LLMs are now being used to automate workflows and build AI Agents. If you are a software engineer, understanding the internals of AI systems is essential. On 28th February, we are launching our first AI Engineering Cohort. A focused 16-week program for software engineers who want real, production-grade AI skills. More details coming soon.

List of most important System Design Interview Questions. cdn.interviewready.io/System…

We are running an End of Year Sale at InterviewReady. If you are looking to upskill on System Design or AI Engineering, this is a worthwhile investment. The courses are built for software engineers who want to build workplace capabilities. Link: interviewready.io Cheers!

2025 has been a breakout year for AI. SDEs, PMs, and TPMs have discovered the power of AI in their organisation. Here is the good news: 1. AI solutions like LLM fine-tuning and RAG pipelines are now mature and production-ready. 2. Newer solutions like agentic systems are gaining traction. 3. AI Engineering has emerged as a dominant future career path. The bad news: 1. There is a massive gap between market demand and available talent (the money is here!). 2. This is still a greenfield with software teams from different organisations wasting effort in building the same solutions. 3. Hiring is picking up, but companies are cautious about overspending in a fast-moving space. Bottomline: The best way to future-proof your career is to learn AI properly in 2026. I went through hundreds of research papers and real-world systems to condense the most valuable lessons into a single, practical course focused on AI Engineering. AI Engineering Course: interviewready.io/course-pag… Wishing you a happy and prosperous 2026. Cheers!

RL for Self-Improving Agents with Skill Libraries 1. Today’s LLM agents don't improve after deployment. 2. When executing a task, they run a sequence of actions. 3. These sequences are stored as a "skill", similar to function calls. 4. Skills are rewarded if they help future tasks succeed. 5. Result: higher success rates, fewer steps, and ~60% fewer tokens used. arxiv.org/abs/2512.17008

Amazon S3 stores exabytes of data across millions of disks. This video explains it's internal architecture. In this video, we break down: 1. How S3 uploads files using parallel multipart uploads 2. How checksums and monitoring prevent data corruption 3. How S3 survives disk, shard, and even region failures 4. How shuffle sharding and request cancellation keep latency low 5. How auto-scaling avoids hot prefixes 6. How error-correcting codes deliver 11 nines of durability with only ~1.8× storage overhead If you want to understand real-world distributed storage design, this is how S3 does it at scale. Links: aws.amazon.com/awstv/watch/d… youtube.com/watch?v=NXehLy7I… interviewready.io/course-pag… #AWS #S3 #SystemDesign

I wrote an article on Long-Running AI Agents. Breakdown: 1. What Are Long-Running Agents 2. When To Use Long-Running Agents 3. Challenges With Long-Running Agents 4. Design Patterns for Long-Running Agents Link: interviewready.io/blog/build…

A very interesting AI trend: Different LLMs are collapsing into the same space. Weights (UWSH): models share the same subspace. arxiv.org/abs/2512.05117 Representations (PRH): models learn the same vector embeddings. arxiv.org/abs/2405.07987 Behaviors (Artificial Hivemind): models produce the same answers. arxiv.org/abs/2510.22954 I have explained PRH in this video: youtu.be/TgiF20Edqlg Cheers.

Cloudflare did it again. Another global outage. On 5 December 2025, Cloudflare deployed a bug fix to its fleet. The fix was to resolve a known vulnerability in React Server Components. They wanted to increase their WAF buffer from 128 KB to 1 MB. This is where things went wrong. 1. Their internal WAF testing tool didn’t support the larger buffer. 2. So a second change was made to disable the testing tool. 3. This config update went through Cloudflare’s global config system, which did not expect a null result from the testing tool. 4. Null Pointer Exception. This is similar to the root cause of the previous outage on 18th November. Cloudflare is revamping their configuration system. Let's hope the new year brings them luck. If you want to learn the engineering behind distributed systems, join @InterviewReady3. Blog: blog.cloudflare.com/5-decemb…

How to Build an AI Agent. I have tried to condense the lessons from the blogs at Anthrophic, Github, and Docker here. interviewready.io/blog/how-t… Cheers!

The best papers to learn AI. github.com/InterviewReady/ai…

The PR is ready. We are moving from the lifetime access plan to one year access. This change will apply to the system design course. Existing users will not be affected by this change. 6 hours to go :D interviewready.io/course-pag…

Changelog of the System Design course: 2020: Launched v1. 2021: Added lessons on HLD and LLD, free of cost to existing users. 2022: Started bi-weekly System Design classes, free of cost to existing users. 2023: Added low-level design course, free of cost to existing users. 2024: Updated System Design Judge, free of cost to existing users. 2025: We are changing our lifetime access plan to yearly access, with 𝗻𝗼 𝗶𝗺𝗽𝗮𝗰𝘁 𝗼𝗻 𝗲𝘅𝗶𝘀𝘁𝗶𝗻𝗴 𝘂𝘀𝗲𝗿 𝗽𝗹𝗮𝗻𝘀. On 27th October, we will be moving System Design to a yearly access plan. All purchases before the deadline will have the lifetime validity plan. Thank you for your continued support. Cheers from InterviewReady. interviewready.io/course-pag…

System Design Course Update. After 26th October 2025, we will shift the System Design course from lifetime to yearly access. The price of the course will remain the same. Currently, we have more than 27000 engineers who use the course. All existing users (who have purchased before 26th October) lifetime validity plan will not be affected. Thank you for your support. Course Page: interviewready.io/course-pag…