Simula

Login with a magic link. Powered by Supabase Auth
Dashboard
Simulation setup screen
Report screen
Credits topup modal

Inspiration

Hackathon teams and early-stage founders are flooded with ideas yet starved for evidence that anyone actually cares. Traditional user interviews are gold, but organizing them can eat up the entire 48-hour sprint. Inspired "Position: LLM Social Simulations Are a Promising Research Method" and related article by a16z, we set out to compress days of research into minutes with AI-powered personas.

What it does

Simula lets you paste a raw product idea, pick (or auto-generate) target personas, choose how many facilitator follow-ups to run, and instantly receive:

A transcript of each AI persona’s reactions
An auto-summarized verdict on which hypotheses were validated
Confidence badges for every response

Key modes:

Realism – each persona secretly starts skeptical, neutral, or enthusiastic to avoid echo-chamber praise.
Self-Consistency Sampling – three parallel answers are merged and weighted by confidence.
Model Picker – GPT-4.1 for depth, GPT-4o-mini for bulk tests, 3.5-Turbo in free demo.

How we built it

Bolt.new for rapid UI prototyping and supabase edge functions Pure OpenAI API calls in TypeScript (Bolt rejects heavier JS frameworks) Hidden “stance” + temperature-fork logic for realistic pushback Credit system or bring-your-own OpenAI key; cost scales with model and settings Deployed to Netlify with the free domain simula.online (Bolt builder pack) React Native app generated from the web codebase in a single prompt

Challenges we ran into

AI sycophancy – early boards only said “Sounds awesome!”; Realism + Sampling fixed this. Bolt constraints – no native Python and picky about libraries, so we kept the stack ultra-lean. Pricing balance – had to offer a free demo without letting users burn GPT-4 credits.

Accomplishments that we’re proud of

Shipped in 6 days, not weeks —from an idea to simula.online.
First dog-food sessions surfaced real security and pricing concerns, proving value.
React Native port produced in one prompt thanks to clean architecture and Bolt.new
In-app monetization and flexible model selection already live.

What we learned

AI personas can save days of research—but only if you inject dissent and measure confidence.
Simple heuristics (stance, sampling) dramatically boost response realism.
Minimal dependencies make deployment on hackathon platforms far smoother.

What’s next for Simula

Template library of ready-made boards for e-commerce, SaaS, gaming, and more.
Dissent 2.0 – a generative “devil’s advocate” engine for deeper critique.
Exports & integrations – push findings to Notion, PowerPoint, or product-analytics APIs. Ground your user boards using your data
Team workspaces with comments, decision tracking, and shared credit pools.
Mobile launch on React Native for on-the-spot interviews at conferences and meetups.