Guidance and project definition. No technical background is required. This is not a promise of a finished software build.
Jai Bhagat software engineer + teacher
10+ yearssoftware and teaching
4.9 / 5.0instructor rating
HashiCorpproduction engineering
Parsonsadult instruction
You do not need more AI homework
You need one decision that makes next week easier.
"I keep trying AI tools, but I still do not have one workflow I trust."
"I have a folder, repository, or recurring task. I do not know what AI should do with it."
"I can make a demo. I do not know whether it is safe, useful, or worth maintaining."
Your session artifact
Leave with a one-page plan.
You can use it on your own, hand it to a technical partner, or bring it into a later build session.
AI PROJECT PLAN30 DAYS
Turn weekly research notes into a reviewable brief
The job
Collect the source notes, group the recurring themes, and draft a brief for human review.
The tool
Start with the paid assistant you already use. Test local AI only if the files cannot leave the machine.
The boundary
The model may read the selected folder. It may not publish, send, or delete anything.
The test
Run four weekly briefs. Track time saved, missing sources, and edits needed before sharing.
A real next step, not a tool list
Hi, I'm Jai
I work where software, teaching, and public service meet.
I spent more than ten years building software and teaching in New York City. I shipped production software at HashiCorp, taught JavaScript at Parsons, and helped Queens founders and operators use AI in work they already care about.
For years, much of my economic empowerment teaching was free. I am now using paid guidance to support the time it takes to publish working demos, share what fails, and build toward a nonprofit collective by June 2027.
My role is not to make every AI idea sound good. It is to help you find the useful one, put a boundary around it, and know what would count as proof.
We choose a recurring task and the result that would matter.
02
Choose the smallest useful test
We narrow the idea until you can run it without rebuilding your whole workflow.
03
Set the tool and data boundary
We decide between local AI, a paid subscription, an API, or ordinary code.
04
Write the 30 day plan
We define what to try, what to measure, and what would make you stop.
Why Shakti is here
The public demo proves when not to use AI.
Shakti reads public NYC housing repair records. Its live lookup uses ordinary code because address matching, record counts, privacy treatment, and public next steps should stay deterministic.
The optional local edition gives AI a smaller job. It explains a finished, treated packet and exposes the tool trace. It cannot change the public record.
