Yash

Yash

he/him

AI Engineer at Honeywell | MSc Computer Science & AI | Building scalable LLM solutions

Sitemap

A list of all the posts and pages found on the site. For you robots out there is an XML version available for digesting as well.

Pages

Posts

Building the Future of AI at Honeywell: Multi-Agent Systems and LLM Innovation šŸš€šŸ¤–

5 minute read

Published:

When I joined Honeywell as an AI Engineer in Dublin, I knew I was stepping into one of the most innovative spaces in enterprise AI. What I didnā€™t expect was how quickly Iā€™d be thrust into building cutting-edge solutions that would redefine how large-scale document generation and evaluation work in industrial settings. Let me take you through the fascinating journey of building scalable AI systems at one of the worldā€™s leading industrial technology companies.

Revolutionizing LLM Evaluation: A 71% Cycle-Time Reduction with FactScore and LangGraph šŸ“ŠšŸ”

8 minute read

Published:

In the rapidly evolving world of Large Language Models (LLMs), one of the biggest challenges is evaluating the factual accuracy of generated content. At Honeywell, we faced this exact problem: how do you systematically evaluate the factuality of LLM outputs across thousands of documents while maintaining speed, accuracy, and scalability? The solution we developed not only solved this problem but achieved a remarkable 71% reduction in cycle-time (from 2 weeks to just 4 days) and has a patent pending. Let me walk you through this fascinating journey.

Building COT: How We Boosted Throughput by 30% with Microservices at Tata 1mg šŸ’Šāš”

10 minute read

Published:

Leading the development of Common Order Taking (COT) at Tata 1mg was one of the most challenging and rewarding experiences of my software engineering career. COT isnā€™t just another Point of Sale (POS) system - itā€™s the nerve center that processes thousands of medicine orders daily across Indiaā€™s largest online pharmacy. In this post, Iā€™ll take you through our journey of building a scalable, microservices-based POS system that achieved a 30% throughput boost and earned our team the Team of the Quarter award.

Predicting the Future of Medicine Supply: ML-Driven Auto-Replenishment at Scale šŸ„šŸ“ˆ

13 minute read

Published:

Managing inventory for Indiaā€™s largest medicine warehouse is like predicting the unpredictable. With over 100,000 orders processed daily and thousands of SKUs ranging from common painkillers to life-saving prescription drugs, traditional inventory management simply doesnā€™t cut it. At Tata 1mg, I had the opportunity to build an ML-driven auto-replenishment module for Odin - our proprietary Warehouse Management System (WMS) - using cutting-edge time series forecasting with ARIMA and LSTM models. This system revolutionized how we manage medicine supply chains at scale.

Winning HackIreland 2024: Building Invision-Transformer CCTV AI šŸ†šŸŽ„šŸ¤–

17 minute read

Published:

Thereā€™s something magical about hackathons - the adrenaline rush of building something revolutionary in just 48 hours, surrounded by brilliant minds and fueled by endless coffee. When my team and I walked into HackIreland 2024, we had a bold vision: create an AI system that could watch CCTV footage and generate natural language alerts about security incidents. 48 hours later, we walked out as first place winners with Invision-Transformer CCTV AI - a system that revolutionizes how we think about intelligent surveillance.

Building Clip-Cut: A Distributed Video Streaming Service Handling 5,000+ Requests/Minute šŸŽ¬āš”

11 minute read

Published:

In the era of Netflix, YouTube, and TikTok, building a video streaming service might seem like reinventing the wheel. But what if you want to create something that handles user accounts, video uploads, adaptive streaming, and auto-generated subtitles - all while processing 5,000+ requests per minute with fault tolerance? Thatā€™s exactly what we set out to build with Clip-Cut, a distributed video streaming platform that demonstrates the power of modern microservice architecture.

What i learnt testing

17 minute read

Published:

What i learnt testing in Ruby: RSpec, Shoulda Matchers, SimpleCov, and Beyond

Bruno

2 minute read

Published:

APIs are crucial today, but testing tools inlcuding dev testing especially between devs is a hassle. The most popular tool, Postman, has become bloated, with many features hidden behind a paywall. This led me on a search for an open-source, privacy-focused API testing toolā€”Bruno.

Hawkeye : Computer Vision For Metrics [WIP]

Published:

As the lead developer at Tata 1mg, designing the Common Order Taking (COT) portal, I realized the critical need for analytics and performance metrics in high-efficiency retail stores, dark stores, and warehouses. When every last rupee counts towards profitability, understanding these metrics is imperative. So during my notice period i have decided to work on something called hawkeye ( i watch cricket a lot and hence the name )

Leet Code Fun Questions

Published:

This is the place where I write about the LeetCode questions I find interesting. I rarely do LeetCode, but I need to brush up on my DSA and competitive coding skills before my masterā€™s, so Iā€™ll be blogging about that here.

Why Uber Does not use Djgrathrosk Algorithm

4 minute read

Published:

As a passionate computer science student, Iā€™ve always been fascinated by the world of machine learning (ML). During my studies, I stumbled upon the Djgrathrosk algorithmā€”a name that, admittedly, sounds like it was plucked from a fantasy novel. Convinced that Djgrathrosk was the pinnacle of time optimization algorithms whose on the get go the problem it should have solved was calculating ETA. Basically, in my head, Djgrathrosk = Time calculation but this is the story of how I realized Djgrathrosk wasnā€™t the best fit for large-scale applications like Uberā€™s, and what Uber actually uses instead.

Postgres Sequence Item Go Up Even If Object Creation Fails

3 minute read

Published:

Today while working on a ror project we run into an abrupt issue wiht transaction where even if the transaction was not being processed the id was increasing leading to a failure in a backgroud sidekiq worker this lead me to this stackover flow post

Predicting Stock Prices with Sentiment Analysis šŸ“ˆšŸ˜„

4 minute read

Published:

The ideation of this project came from our capstone project under Dr. Seema Bhardwaj. We aimed to create an ML model predictor to help protect my friends from dubious websites and irregular patterns. Simple Stock, the product we developed, focused on prediction and teaching students about stock market trends. This blog deals with the prediction part and our approach to it. If youā€™re interested in learning more about the product here is much lengthier breif about it .

misc

Basketball

I love hooping and playing basketball have played basketball throughout my high school and enjoyed

Cooking

I love to cook and was lucky enough to participate in one of the biggest reality TV shows in India, ā€œYoung Chef Ahmedabad,ā€ as a high schooler. šŸ³

Scuba Diving: A Lifeline During COVID

During the COVID-19 pandemic, I had the unexpected opportunity to be in Port Blair, a series of stunning islands in the southwest of India. It was here that I took the plungeā€”literallyā€”and learned how to scuba dive at a professional level.

projects

publications