One of the things I love doing is bringing ideas to life from actual research papers — combining evolutionary strategies with simulation and pulling ideas from papers most people never read.
When I use AI to turn a paper into a working simulation, there are two options I have to consider:
1. Write the plan myself from what I know and keep iterating until it's right.
2. Let the AI read the papers and plan the features from them.
The best way is to combine both. So that's exactly what I did.
I built a research warehouse (datapilot.justgotalead.com) that turns what you know into plans your AI can build from.
You put in the papers, books, repos and documents for a project. It reads them, and answers questions from them with citations so you're not asking an AI what it remembers, you're asking your own library.
When you type an idea, it plans the whole build from that material: the screens, the data model, the goals, each step naming its source.
Then your own AI builds from that plan, and it stops making up answers, because context no longer has to come from you. Your AI asks your own research library how to build the thing.
Fewer hallucinations, less build time.
I ran the same idea twice. Once with it, once without.
"Build me a 3D first person shooter game" straight into Claude Code: 2 hours, 150k+ tokens.
Same prompt, built from a DataPilot plan: 1 hour, ~50k tokens.
Ultimately, we are separating the planning and the building, so you save tokens and put your best model in building.
My goal is to have other users test this software out for free because we want to continue to improve this and hopefully allow more feasibility for users to spend less time planning and more time building higher quality products.
One of the things I love doing is bringing ideas to life from actual research papers — combining evolutionary strategies with simulation and pulling ideas from papers most people never read.
When I use AI to turn a paper into a working simulation, there are two options I have to consider:
1. Write the plan myself from what I know and keep iterating until it's right.
2. Let the AI read the papers and plan the features from them.
The best way is to combine both. So that's exactly what I did.
I built a research warehouse (datapilot.justgotalead.com) that turns what you know into plans your AI can build from.
You put in the papers, books, repos and documents for a project. It reads them, and answers questions from them with citations so you're not asking an AI what it remembers, you're asking your own library.
When you type an idea, it plans the whole build from that material: the screens, the data model, the goals, each step naming its source.
Then your own AI builds from that plan, and it stops making up answers, because context no longer has to come from you. Your AI asks your own research library how to build the thing.
Fewer hallucinations, less build time.
I ran the same idea twice. Once with it, once without.
"Build me a 3D first person shooter game" straight into Claude Code: 2 hours, 150k+ tokens.
Same prompt, built from a DataPilot plan: 1 hour, ~50k tokens.
Ultimately, we are separating the planning and the building, so you save tokens and put your best model in building.
My goal is to have other users test this software out for free because we want to continue to improve this and hopefully allow more feasibility for users to spend less time planning and more time building higher quality products.
[dead]