Recent content on Sumit Gouthaman
I was recently using Q CLI to implement a new feature and noticed it struggling in an unexpected way. It had managed to implement the feature incredibly quickly in 4-5 iterations but was struggling when it came to updating all the tests impacted by the change. In particular, it was repeatedly running into issues editing large text files either because the string it was trying to replace…
LangGraph is a framework I appreciate for its ability to abstract away the underlying LLM provider. However, when it comes to multimodal messages, the underlying implementation details leak through.
How I extract useful structured data from messy brokerage statements.
Imagine this: You’re building some kind of a chat system. For every message from the user, you want to figure out what the message is about. Is the user asking a question? Or does the user asking to perform a specific action? Or is the user talking about something completely irrelevant to your system? What we are doing here is detecting the “intent” of the user’s message.…
I’ve recently been spending some time improving my understanding of distributed systems. YouTube is now the de facto way I learn new things, so that’s where I naturally started. There are definitely some great channels on YouTube (see my recommendation list below), but the vast majority of videos related to this topic on YT are quite terrible. In general, they tend to be focussed on…
Are you planning to use Go for solving coding problems in an upcoming interview? If so, you might have realized that the language feels a little “unergonomic” for quickly implementing algorithms under the time constraints imposed by an interview. Unlike the real world, in a coding interview lack of conciseness and excess boilerplate gets in the way of finishing the solution in time and…
Hi, I am Sumit, a Software Engineer based in Seattle, Washington. I’m interested in a wide variety of topics, both software engineering related (distributed systems, Kubernetes, LLMs, developer tools, etc.) and not (biking, cooking, history, politics). This blog is a random assortment of things I find interesting. I’ll post here whenever I come across something I find useful. This is…
Vibe coding experiments Quick projects built while experimenting with various tools and approaches. Podcast Smart Trim 📂 View on GitHub A CLI tool that uses AI to automatically identify and remove advertisements from podcasts. This project was built to test Antigravity , Google’s new IDE. Features: AI Transcription : Uses openai-whisper for accurate transcripts. Smart Ad Detection : Uses…