Inspiration
I work with market research teams and see the same problem repeatedly: companies want to test ideas with consumers, but doing so can be expensive and time-consuming. For example, a company might want to know which promotion will drive more protein bar sales: "40% off all bars!" or "Buy one, get one free!" The company will then design an online survey, and pay hundreds of people to take the survey online costing hundreds or thousands of dollars and precious time. They then manually analyze the results and learn the two promotions perform identically. Is there a faster and cheaper way to come to the same conclusion?
Doppelganger was built to solve this problem. We provide market researchers the ability to test their surveys with AI participants prior to spending significant time and money on a real human test. Within minutes hundreds of AI participants will take the survey and give market researchers critical information on their survey. We do not replace running real surveys on humans, but are a helpful aid to "screen out" certain ideas before spending time and money on a real survey.
What it does
You paste a survey link and Doppelganger runs it through hundreds of persona-based AI respondents (e.g. a 19 year old biology student who takes surveys in her free time to earn a little extra money). The industry standard for online surveys is Qualtrics. We use the Qualtrics Sessions API, so real branching and randomization fire exactly like they would for a human. Your survey gets run through three different LLMs in parallel, and you compare results across them.
Credit purchases run through PayPal, new accounts get 200 free credits so you can try it without paying, and AG Grid powers the entire results experience. We use AG Grid to show a respondent-level table where you can sort and filter every individual answer, and a cross-model comparison view that puts all three LLMs' results side by side.
We benchmarked the ensemble against the published Twin-2K-500 dataset: it reproduced 15.5 of 17 classic behavioral effects, versus 9 for the digital twins built in the original study.
How we built it
FastAPI on Railway, Next.js on Vercel, Postgres, Gemini plus two OpenRouter models for the ensemble. PayPal is a server-side Orders v2 integration: the client only ever sends a pack id, prices live on the server, and capture is idempotent so a double click can't double-credit. AG Grid was used to help display the survey results.
Challenges we ran into
Qualtrics rejects AI sessions on lots of question types, so we auto-generate an AI-compatible copy of your survey and never touch the original.
Accomplishments that we're proud of
We have real paying customers! The benchmark results gives us confidence this project is not just a gimmick. Pretesting with AI only matters if the AI behaves like people, and 91% replication of known human effects is a real number from an external dataset, not a vibe.
What we learned
Building real products that serve actual customers is hard, even in this vibe coding age.
What's next for Doppelganger
AI-gated payouts. After you field the real survey, the same models that pretested it score each human response for quality (speeding, straight-lining, gibberish), and legit respondents get paid automatically through PayPal Payouts. Bots don't.
