Shanif Dhanani (@shanif) header
Shanif Dhanani (@shanif) profile picture

Data scientist / software engineer. Building Nobi, Milton, and retail demand planning systems.

Building AI assistants →
Joined April 2009
Gotta love New York
33
1/ I've been working with a company that has gone all in on having their employees vibe code internal dashboards. It all works great, until it doesn't. Employees need software tools and reporting. AI makes it easy for them to create their own.
1
52
4/ 1. Make sure everyone's vibe-coded project has a home on your company's GitHub account 2. Run a simple automated code review process that ensures there are no obvious security leaks (like passwords in source code)
1
16
5/ 3. Use GitHub's actions and runners to keep data up to date without having to run on an employee's laptop 4. Use Vercel or Netlify or even Claude/ChatGPT artifacts for deploying A few simple things go a long way.
17
Zain Yousaf and I just finished our latest Show & Tell. On this week's share session, we went over 3 things: 1. Claude Design for decks 2. Launching multiple Claude apps 3. A new Shop Nobi feature that lets you search across lots of stores Check it out: youtube.com/live/z-ZOSuUuyb0…
44
Another weekly show & tell done with my buddy Zain Yousaf: youtube.com/watch?v=rwFXrMbD… Stay tuned for next week's link, and if you're interested in joining us, let me know!
105
Another Friday Show & Tell completed with my buddy Zain Yousaf. We go live every Friday at noon ET to talk about what we're seeing and doing. Come join us! Next week's URL coming soon. youtube.com/watch?v=010aTFOh…
1
1
41
This week was all about demand forecasting, AI-native Slack replacements, and chatting with AI.
12
1/ I've added better explainability to Milton's predictions. The stock forecasts I was seeing were unclear as to why they were what they were.
1
71
2/ This explainability (which is a major area of research in ML and one that, I believe, is likely custom to each specific scenario) helps me understand WHY a particular stock forecast is what it is.
1
36
3/ Milton is now out there at milton-nau-web.pages.dev/ (but at some point I need to think of a better name and URL).
32
1/ One area where I've just been hitting a wall with AI coding agents is in their ability to complete, end-to-end, a highly complex coding task without any human input.
2
2
74
3/ Dealing with this problem has become time-consuming enough that I am considering building some sort of harness/overseer/guide that can ensure a task is completed satisfactorily to my liking.
1
16
4/ But that would be a huge task and might even require a lot of new techniques/technologies that have yet to be discovered. I can't be the only one running into this. Has anyone found any solutions for this problem?
6
1/ Over the pst couple of months I've built out a demand forecasting system that can help retailers identify which products they should be buying/planning for.
1
55
9/ If you're not doing that yourself, it's a great time to start.
1
1
4
10/ You likely have a ton of operational data that you can use to build in automations, and now that we have LLMs to help us create those systems, it's never been easier to start using your data to improve your business.
4