Rahul Roy Chowdhury

CS’25 | Full Stack Developer | Always learning

[Rahul Roy Chowdhury @Rahulstark4]

didn’t know I’d be this interested in video engineering

https://t.co/TqKelNu0Gw

[Rahul Roy Chowdhury @Rahulstark4]

The likely culprit is a missing idempotency mechanism. The payment succeeds, but the response times out. The backend retries and charges the customer again. The database records only one payment while the gateway records two. Use idempotency keys to prevent duplicate charges. Always verify payment status before retrying.

[Rahul Roy Chowdhury @Rahulstark4]

Your payment API has a 99.99% success rate.

Your database shows 12,000 successful payments.

Your payment gateway shows 12,347.

No errors in your logs.

Where did the money go?

[Rahul Roy Chowdhury @Rahulstark4]

Great read 🙌

[Naman Jain @namanjn1998]

I think the biggest gap in voice AI is that agents make decisions from a transcript, not from the audio. I wrote down my thoughts.

I’d love to hear where you disagree. https://t.co/hHmfjlbFKG

[Rahul Roy Chowdhury @Rahulstark4]

reverse engineering is becoming a lot less of a barrier

once models can recreate almost any software by analyzing it, keeping your code private might not be the competitive advantage it used to be

[Rahul Roy Chowdhury @Rahulstark4]

check this out https://t.co/KdTouX8GXd

[Rahul Roy Chowdhury @Rahulstark4]

sometimes the best way to learn something is to stop watching tutorials and start building something from scratch. you understand things very differently when nothing works and you have to figure out why.

[Rahul Roy Chowdhury @Rahulstark4]

@Ray99Bimal It’s object

[Rahul Roy Chowdhury @Rahulstark4]

had it bookmarked for ages btw… finally got around to it

https://t.co/9Mb5frvBXB

[Rahul Roy Chowdhury @Rahulstark4]

wait who the hell was controlling the AI

[Rahul Roy Chowdhury @Rahulstark4]

My first Rust PR just got merged!

Contributed to YC backed @agent_relay by fixing how host Rustup Cargo shims are handled in Relayflows.

Learned a lot from this 😌

[Rahul Roy Chowdhury @Rahulstark4]

how to decide when to use a monolith and when to use microservices ??

This is a really broad question and the answer depends on your system’s workload characteristics, domain boundaries, team topology, and operational maturity.

I’ll try to share some general thoughts, as someone who is still learning, but every architecture comes with different trade offs, and there is no one size fits all solution.

A monolith makes sense when you’re building an early stage product with a relatively stable domain model. A modular monolith gives you strong transactional consistency, simpler debugging, and lower inter-process communication overhead without introducing unnecessary distributed systems complexity.

Microservices become more attractive when bounded contexts have distinct scaling requirements, independent deployment lifecycles, or different availability and fault isolation requirements. You can scale individual services independently, but you also introduce network latency, partial failure scenarios, eventual consistency, and distributed tracing challenges.

One thing people often underestimate is the complexity of distributed systems. Service discovery, circuit breakers, idempotency, retry policies, distributed transactions, and observability aren’t optional concerns once your architecture starts spanning multiple services.

A monolith isn’t inherently a scalability bottleneck, and microservices don’t automatically improve throughput. Poorly designed service boundaries can introduce chatty communication patterns, cascading failures, and excessive serialization overhead. Sometimes the actual bottleneck is database contention, inefficient query execution, or lock contention rather than the application architecture itself.

Infrastructure cost matters too. Container orchestration, service meshes, CI/CD pipelines, centralized logging, and telemetry aggregation all introduce additional operational overhead. Without sufficient platform engineering maturity, the cognitive load can outweigh the benefits of independent deployability.

Personally, I’d start with a modular monolith, establish clear domain boundaries, and use profiling, workload characterization, and performance metrics to identify genuine bottlenecks before extracting services.

Architecture should evolve around measurable constraints and business requirements, not hypothetical scaling problems.

[Rahul Roy Chowdhury @Rahulstark4]

today’s paper read is about algorithms, optimization, and the limits of efficient computation

https://t.co/xwWnZUceU8

[Rahul Roy Chowdhury @Rahulstark4]

The interesting part isn’t that Anthropic built its own Git.

It’s that AI is pushing existing engineering infrastructure beyond the limits it was designed for.

When the volume of code changes grows this fast, the bottleneck shifts from writing code to everything around it.

We’re entering a world where our tools need to evolve just as fast as the models using them

[Anthropic @AnthropicAI]

New on our Engineering blog: Creating in-house source control.

In May, we merged around 2,500 code changes per day. In September, we merged more than 30,000 per day. As a result, we became bottlenecked by developer infrastructure built for human (rather than agent) scale.

Here, we share how we built a new version control system—and fully migrated our monorepo from git—in five months: https://t.co/wzNkwHs7IN

[Rahul Roy Chowdhury @Rahulstark4]

Amazing

[Mark Zuckerberg @finkd]

Introducing Muse Wonder, our new family of frontier models with state of the art performance for personal and professional agentic work.

[Rahul Roy Chowdhury @Rahulstark4]

If anyone’s interested in Inference Engineering, this is a great introductory video.

https://t.co/GZsElQtCqC

[Rahul Roy Chowdhury @Rahulstark4]

just read this netflix blog on making their data infrastructure more cost efficient

pretty cool to see how they approach cloud costs at scale. learned a few things i hadn’t thought about before

https://t.co/6VqixkqMsI

[Rahul Roy Chowdhury @Rahulstark4]

AI won’t replace engineers who can think.

But it will expose engineers whose only skill is translating instructions into code.

The real value is understanding the problem, making architectural decisions, and building things people actually need.

Don’t compete with AI at writing code. Learn to solve problems AI cannot define on its own.

[Rahul Roy Chowdhury @Rahulstark4]

India should welcome competition, not protect companies from it.

Data sovereignty matters, but so does giving millions of Indians access to education, opportunity, and a global economy.

Let the best service win 🇮🇳

[Elon Musk @elonmusk]

Dear Prime Minister Ambani,

Please accept my humble apologies for not realizing that you are the real boss of India.

Naturally, you would prefer to maintain your monopolistic exploitation of the great people of India, but would you nonetheless consider allowing Starlink to compete?

There are many areas of India, as there are in all countries, with no Internet access, which denies children the opportunity for education and for small businesses the opportunity to sell their goods to a global market. Starlink would change their lives for the better.

Thank you 🙏