shekhargulati

I also started building similar tools in early 2025 and have built around 80 of them so far: https://tools.o14.ai/. Recently, I built a data viewer app (https://tools.o14.ai/excel-qa-review.html) where I can upload an Excel or CSV file containing chat queries and answers for quick manual review.

Reviewing data in Excel is painful, especially when answers are in HTML or Markdown, because you don’t get proper rendering. Building small, custom tools that reduce the friction of reviewing data makes life much easier and more pleasant. These days, I use Claude Code for Web to build most of these apps, and they are deployed on Vercel.

I tried to port it to Docker and wrote a blog here https://shekhargulati.com/2025/08/09/making-coderunner-ui-wo.... I used Claude Code to do the port. We used Datalayer Jupyter MCP Server instead of coderunner which uses Apple containers.

Not sure why they used React for a CLI. The code in the repo feels like it was written by an LLM—too many inline comments. Interestingly, their agent's system prompt mentions removing inline comments https://github.com/openai/codex/blob/main/codex-cli/src/util....

- Remove all inline comments you added as much as possible, even if they look normal. Check using \`git diff\`. Inline comments must be generally avoided, unless active maintainers of the repo, after long careful study of the code and the issue, will still misinterpret the code without the comments.

Google AI Overview incorrectly identified the day for a given date due to a timezone conversion issue, likely using PST instead of IST. ChatGPT and Perplexity provided more accurate and detailed responses.

I detailed it on my blog here https://shekhargulati.com/2025/03/23/google-ai-overview-has-...

I am building https://www.videocrawl.dev/, an AI companion web application that enhances the video-watching and learning experience. Since I primarily learn from videos, I built this to improve my learning workflow. It is free to use.

It offers standard features like chatting with videos, summarization, FAQs, note-taking, and extracting sources mentioned in the video. We're taking it a step further by extracting relevant information from video frames and making it easily accessible.

We built Videocrawl [1] to enhance the learning and watching experience using LLMs. It handles the usual tasks like clean transcript extraction, summarization, and chat-based interaction with videos. However, we go a step further by analyzing frames to extract code snippets, references, sources, and more.

You can try it out by watching a video on Videocrawl, such as the OpenAI Agent video, by following this link [2]. LLMs have the potential to significantly improve how we learn from and engage with videos.

1. https://www.videocrawl.dev/ 2. https://www.videocrawl.dev/studio?url=https%3A%2F%2Fwww.yout...

Mistral OCR made multiple mistakes in extracting this [1] document. It is a two-page-long PDF in Arabic from the Saudi Central Bank. The following errors were observed:

- Referenced Vision 2030 as Vision 2.0. - Failed to extract the table; instead, it hallucinated and extracted the text in a different format. - Failed to extract the number and date of the circular.

I tested the same document with ChatGPT, Claude, Grok, and Gemini. Only Claude 3.7 extracted the complete document, while all others failed badly. You can read my analysis here [2].

1. https://rulebook.sama.gov.sa/sites/default/files/en_net_file... 2. https://shekhargulati.com/2025/03/05/claude-3-7-sonnet-is-go...

I asked Claude 3.7 Sonnet to generate an SVG illustration of Maha Kumbh. The generated SVG includes a Shivling (https://en.wikipedia.org/wiki/Lingam) and also depicts Naga Sadhus well. Both Grok 3 and OpenAI o3 failed miserably.

You can view the generated SVG and the exact prompt here: https://shekhargulati.com/2025/02/25/can-claude-3-7-sonnet-g...

I’m building Videocrawl https://www.videocrawl.dev/ a platform designed to make video content more accessible and actionable. It offers two key features:

1. API for Clean Transcripts – Extract structured transcripts with references, code snippets, and images.

2. AI Video Assistant – Interact with videos using AI-powered tools for summarization, Q&A, and more.

Would love to hear your thoughts! You can reach out to me at sg@o14.ai

I don't know, but I found the recording uninspiring. There was nothing new for me. We've all seen reasoning models by now—we know they work well for certain use cases. We've also seen "Deep Researchers," so nothing new there either.

No matter what people say, they're all just copying OpenAI. I'm not a huge fan of OpenAI, but I think they're still the ones showing what can be done. Yes, xAI might have taken less time because of their huge cluster, but it’s not inspiring to me. Also, the dark room setup was depressing.

Just a side note: The Wikipedia page for "Deep Research" only mentions OpenAI – https://en.wikipedia.org/wiki/Deep_Research

I wrote a simple LLM tool that helps generate such scripts.

Tool: https://tools.o14.ai/do-nothing.html

Blogged about here: https://shekhargulati.com/2025/02/16/do-nothing-script-gener...

running things in production/self hosting is more annoying than just paying like 20-200/month

This is an important point. As people rely more on AI/LLM tools, reliability will become even more critical.

In the last two weeks, I've heavily used Claude and DeepSeek Chat. ChatGPT is much more reliable compared to both.

Claude struggles with long-context chats and often shifts to concise responses. DeepSeek often has its "fail whale" moment.

Have people tried using R1 for some real-world use cases? I attempted to use the 7b Ollama variant for my UI generation [1] and Gitlab Postgres Schema Analysis [2] tasks, but the results were not satisfactory.

- UI Generation: The generated UI failed to function due to errors in the JavaScript, and the overall user experience was poor.

- Gitlab Postgres Schema Analysis: It identified only a few design patterns.

I am not sure if these are suitable tasks for R1. I will try larger variant as well.

1. https://shekhargulati.com/2025/01/19/how-good-are-llms-at-ge... 2. https://shekhargulati.com/2025/01/14/can-openai-o1-model-ana...

This is the same approach we took when we added LLM capability to a low code tool Appian. LLM helped us generate the Appian workflow configuration file, user reviews it and make changes if required, and then finally publishes it.

I also built something similar using yt-dlp and llm CLI and wrote a post about it https://shekhargulati.com/2024/07/30/building-a-youtube-vide.... Script here https://github.com/shekhargulati/llm-tools/blob/main/yt-summ...

I am author of this post. Let me know if I can answer any question related to the post.

I did something similar in 2016 where I wrote tutorial on new technologies almost every week https://github.com/shekhargulati/52-technologies-in-2016. You learn a lot but as you don't end up using them you forget them .

I will not start with programming but will focus on:

1. Learn how to build a good understanding on a subject. As I am growing old, I feel I am in hurry to grasp more rather than deeply learning stuff

2. Second important area for me will be how to effectively communicate my point point of view and listen to other.

In short, I will work on my soft skills more.

I also ran couple of such series:

2013: 30 technologies in 30 days https://shekhargulati.com/30-technologies-in-30-days/

2016: 52 technologies in 2016 https://github.com/shekhargulati/52-technologies-in-2016

I am author of this series. This series has been a great experience for me to learn new things and apply the learnings in professional projects.

Same here this is a very bad blog. I don't see what you can learn from this tutorial. If you want to learn Java I wrote a series https://github.com/shekhargulati/java8-the-missing-tutorial covering new features of Java 8.

Shameless plug: I am also doing something similar https://github.com/shekhargulati/52-technologies-in-2016. I am learning a new technology every week or writing some small open source project mainly in Scala, Python, and Java. I am documenting my learning on Github.

It depends from 5 hours to 15 hours depending on the topic. On Saturday I learn a topic by reading documentation or watching videos tutorials by their creators(if I can find) or if I have a book then I read first few chapters. Sunday, I start with building something and then documenting the process in the blog by end of Sunday.

I am doing a series called 52-technologies-in-2016 https://github.com/shekhargulati/52-technologies-in-2016 where in I learn a new technology, build a small app, and blog about it every week. I maintain an Evernote where I write down all the interesting topics or projects I find. I go through the list and randomly pick any topic that excites me that day and then work over the weekend to publish something. This helps me keep in continuous learning loop.

Last nine weeks covered Finatra, SBT, Slick, REST API Clients using OkHttp, Hugo, CoreOS, and Google Cloud Vision API.

I have used Hugo and I found it more performant and easy to use.

So sorry for that. I have changed the theme let me know if that helps.

I have taken a personal challenge to learn a technology every week and build an application using it. So far I have written about Scala covering SBT, Finatra, Slick, performing sentiment analysis in Scala using Stanford CoreNLP. Let me know what you all think.

Yes, you have to poll CodePipeline every second or so to get the jobs and perform your task. Currently, web hooks are not implemented or they are WIP.