AI writes the code.
Engineers write why.

A clearer understanding of how to present my projects in interviews. The technical work matters, but so does explaining the reasoning behind it. Postponed practicing DSA problems today. Need to make that a non-negotiable habit

My First Log: Today I want to share a lesson I learned when I developed the MVP of my research project: a Self-Reflective RAG platform designed to detect low-confidence retrievals and decide when it should search for more context before answering. The idea sounded straightforward. If retrieval quality is poor, expand the search. More context should mean better answers. Right? So I built a Heuristic Guardrail Engine that measures retrieval quality before the LLM ever sees the context. If the risk score is high, the system triggers a self-reflection loop and expands the search window automatically. Then I benchmarked it. 50 test queries: In-domain questions, Multi-hop questions and Out-of-domain hallucination traps. And that's where things got interesting. The self-reflection loop worked exactly as intended for low-confidence in-domain queries. It rescued answers that would otherwise have failed. But on purely out-of-domain queries, the opposite happened. By expanding the search, I was feeding the model more irrelevant information. More context didn't create more certainty. It created more noise. And sometimes that extra noise increased the chance of a hallucination slipping through. WHY? Because I assumed every low-confidence retrieval problem could be solved the same way: retrieve more context. The benchmark proved that assumption wrong. Some queries are low-confidence because the answer exists but wasn't retrieved well. Others are low-confidence because the answer doesn't exist in the knowledge base at all. Treating both cases identically makes the system worse, not better. The most valuable lesson from this project wasn't about RAG. It was about learning when two problems that look similar actually require completely different solutions. I started with: "Low confidence → retrieve more." I finished with: "First determine whether the problem is low confidence or out-of-domain." GitHub: https://github.com/Pandidharan22/Heuristic-Self-Reflective-RAG Live Demo: https://self-reflective-rag.pandidharan.dev/

Traced how my API Gateway handles JWT today. The gateway receives the request, pulls the token from the Authorization header, verifies the signature using the secret key, and only then forwards the request to the right microservice. If the token is invalid it never reaches the service at all. I wrote this code months ago but I was thinking of it as "authentication middleware" — today I understood it's actually a bouncer at the door, not a check inside the room. The decision to verify at the gateway level instead of each service was the right one — because if each service verified tokens independently, a bug in one service's auth logic could expose all the others. One gate is easier to secure than six. I just didn't know why until now over each service checking its own token because one gate is easier to secure than six

over because I had both docker-compose.yml and render.yaml in my e-commerce repo. I thought they were alternatives — pick one. Today I realised they solve different problems. Docker because I needed all 6 services to run identically on my machine without "works on my laptop" issues. Render because I didn't want to manage servers in production. The decision was always there — I just never named it while buildin

An Multi Agent web application with RAG Mechanism using langchain framework. It basically used to manage all the cyber security threats and remediations to terminate them(under developmentGit hub: https://github.com/MITHULESH07/AI-powered-cyber-security-thread-intelligence-systemWeb app link: https://ai-powered-cyber-security-thread-in.vercel.app/

not to support string "True" as a value for a key in the payload in one of our APIs over proper JSON true because that's sloppy API design that I don't want me or future colleagues to support.

the regressions we're introducing and follow up fix PRs we're deploying as part of our  moving fast because I'd like us to move fast and be correct rather than moving fast and breaking things.

I didn’t know any programming at first. Then, on 16/11/2024, my birthday, I met a man named Kaniraja who was working at Mercedes-Benz Group. I was interested in the automotive industry. A few days later, I called Kaniraja, and he gave me some ideas and a problem related to a bus tracking system. I started from nothing, but I had a clear goal: to solve the student bus tracking problem. Everything was new to me — new programming languages, new frameworks, IoT devices, and many other things. Every day, I learned something new and applied it to my project. I didn’t do it just as a college project. I genuinely wanted to solve a real-world problem. In my surroundings, many people don’t even know simple things like changing passwords, recovering accounts, or managing Google or Facebook accounts properly. Even in that environment, I built an application to solve a real problem. After around 15 months of hard work, I completed the project. But my college completely ignored it.”

I learned one more valuable things from my mentor & my boss and also my well wisher - He always told me - Before touching figma, understand the problem context and then suggest me to work on UI part next. The exact mistake, I repeated a lot because of deadlines and being in some complex situations. Unfortunately, in the end - the mistake addressed sharply by my boss. Today, I realized and corrected my mistake, work for the value & the impact I create - not for the submission or working like you don't like the work.

Product thinking goes hand-in-hand with customer context because you cannot decide (or delegate to an agent) what to work on without knowing what to build.

I learned that marketing and branding are understood better through real conversations, live brainstorming, and practical discussions rather than only watching videos or reading blogs from friends who are already working in the content and branding field.

an issue within our system where deploys make long-running jobs lose state, there are a couple of solutions that we can try but for that I need to understand and verify how exactly deploys work and replace running code in production.

G Surendar Thina for For giving me an opportunity to work with him at YCS.That experience made me go deeper into marketing and helped me gain more confidence and exposure.It also made me think more independently, take ownership, and understand how real work actually happens.Because of this journey, I’m learning skills and experiences that I believe will help me build my own company in the future.

My mentors & friends for being a mirror to see stuff I miss seeing in myself and being a wall I can bounce of my ideas off of.

Focus more seriously on founder branding/content ecosystem instead of blindly following placement prep over Continuing full placement-focused Java/corporate prep path... because I realized my natural strengths align more with communication, positioning, storytelling, and working with founders than pure repetitive coding roles

Wednesday I got stuck because of some minor coding errors.1. I used "=" instead of "==" in the if condition.   "=" assigns a value, but "==" compares values.2. I wrote "digitalwrite()" instead of "digitalWrite()".   Arduino is case-sensitive, so the capital "W" is important.

I decided to start the backend and aws learning over just working on flutter because for wide range of knowledge and it'll be helpful for standalone projects

You don't need a physical location to build attention and trust in your customer, meet them where they are from Seth Godin's blog titled "In defense of popups"

Another thing - critique & feedbacks for the my designs. When ever, I explaining the designs with the developer team, I will face a lot of good questions & points, I missed out. I learned this - Collaborating with founders & dev folks - would open a door for more clarity on the process. Improves the quality of the outcome. Importantly, the burnout decreases. Expectations matched perfectly.

I learned why small annotations and micro-interactions are so important when presenting designs to stakeholders and founders. from What I’ve learned from my failures.

I selected the Hybrid Test Framework (POM) over the Modular Framework because it provides better scalability, reusability, and maintainability for automation testing projects. By using the Page Object Model (POM), web elements and test scripts are separated, making the framework easier to manage and update when application changes occur. The hybrid approach also combines the advantages of data-driven and modular testing concepts, improving test efficiency and reducing code duplication. This framework is highly suitable for large and complex applications where stability, readability, and faster test execution are important.

Today I revamped the authentication system to make it production-ready. I moved JWTs into HttpOnly cookies for better security, added CSRF protection, built custom login/logout logic, also verified the full flow from registration to logout.

I made a decision to work on the priority task of complementing the mid-fidelity wireframe first and planned to connect with manager during usability testing.

Gratitude because Not being gratitude makes me stressed

How product is to be built because I understand incentives of people now

I thinking about, today i planned to study but I didn't do that according to my plan because I get distracted by few things like using mobile and didn't focused on my studies and most importantly i didn't curious about that

proof app