What is Jahan?
Jahan is a research instrument for simulating how a population might respond to a product. It creates LLM personas grounded in real survey respondents, places them in a social network, and exposes them to product discussions through an X-like feed, a Reddit-like forum, and word of mouth; personas can react, talk, shift views, and answer survey waves about purchase intent. A local web interface supports setting up and running studies, then reviewing replays, reports, and traces. Results are uncalibrated, and the separate persona corpus is research-only; commercial studies require a population users have rights to.
Jahan Features:
Population Grounding: Jahan builds its personas from real survey respondents rather than inventing them from a prompt; the underlying corpus is separate from the repository.
Social Network: Personas are placed in a network where they can encounter one another and pass word of mouth along their ties.
Product Channels: Product exposure happens through an X-like feed ranked by network proximity, similarity, and age, a Reddit-like forum ranked by hotness, and word of mouth.
Survey Waves: Personas answer successive survey waves about whether they would buy a product, with responses expressed in their own words.
Intent Scoring: The engine scores purchase-intent answers by semantic similarity to frozen anchor statements rather than asking personas to provide a number.
Recorded Traces: Each turn is stored in an append-only trace that supports replay, resuming, and auditing; prompts can be reconstructed and checked against their recorded hash.
Trust Reporting: Reports state that trust is uncalibrated, and a published holdout evaluation found filled-in attitudes performed worse than a plain demographic baseline.
Jahan Benefits:
Inspectable Results: Tracing a reported number to the turn that produced it gives researchers a way to check how an outcome was formed.
Repeatable Runs: Replay and resume support makes it easier to revisit a simulation without relying on a written summary of what happened.
Contextual Reactions: Responses shaped by feeds, forums, and social ties provide more context for exploratory product research than purchase-intent answers alone.
Clear Limits: Visible trust caveats and published negative findings help teams avoid treating simulated purchase intent as validated market evidence.
Research Workflow: Keeping study setup and output review in one local interface gives researchers a direct route from defining a population to examining findings.
Use Cases:
Product Concept Testing: Introduce a proposed product to a simulated population and review how personas react, discuss it, and respond in survey waves.
Purchase Intent Research: Run survey waves to examine how personas answer whether they would buy a product after encountering it in the simulated world.
Social Exposure Studies: Explore product reactions when people encounter it through an in-network feed or a Reddit-like discussion forum.
Word of Mouth: Study how product discussion could travel between connected personas through the network's social ties.
Study Auditing: Inspect the trace behind a reported result when a researcher needs to see which turn produced a particular number.

