# AIDEA — AI Idea Validation (Full Reference) > This is the comprehensive version of llms.txt for AIDEA. > For the concise version, see: https://aidea.life/llms.txt > Last updated: 2026-05-28 AIDEA is a mobile app that validates your idea with 100 NVIDIA Nemotron personas, delivering 8 independently scored metrics — understanding, need fit, curiosity, trial intent, repeat intent, pay intent, trust concern, and switching friction — plus risk hypotheses, a validation plan, suggested interview questions, and raw persona responses grouped by sentiment. It is designed for solo founders, product managers, and designers who need to quickly test an idea direction before committing to real user research. --- ## What AIDEA Does AIDEA is an idea validation app built for the earliest stage of product thinking — the moment when you have an idea but no one to ask yet. You write your idea in one paragraph, specify a target audience and their situation, and AIDEA runs it through AI personas selected from the NVIDIA Nemotron-Personas dataset (7 million+ records, CC BY 4.0 license by NVIDIA). Each persona responds independently from their own demographic and psychographic perspective. The raw responses are then deduplicated, aggregated with weighted scoring, and structured into a detailed validation report. The core philosophy is honesty. AIDEA surfaces weaknesses and risk hypotheses first, not just positive signals. It positions itself as a "first mirror" — a tool for checking your idea alone before talking to real users. It does not claim to replace user interviews, predict revenue, or guarantee success. The app is bilingual (Korean and English) and currently available on iOS, with Android planned. --- ## Quick Validation (v0.2.0) Quick Validation is the primary input mode introduced in version 0.2.0. It reduces the idea input process from four separate steps (idea type, target audience, pain points, detailed description) to a single sentence. ### How Quick Validation works Instead of completing four separate forms, you write one sentence describing your idea. AI then automatically fills in the structured fields that used to require manual input. ### Four sub-features of Quick Validation 1. **AI auto-estimation** — When you type your idea sentence, AI automatically detects and fills in the idea type, target audience, geographic region, and age group. You can review and adjust any of the auto-estimated fields before starting validation. 2. **Clarity meter** — A real-time indicator that shows whether your input contains enough information for meaningful validation. It updates as you type, so you know at a glance when the input is ready. This prevents users from submitting vague one-word ideas that would produce low-quality results. 3. **AI follow-up questions** — If the clarity meter indicates the input is incomplete, AI generates targeted questions that pinpoint exactly what is missing. These questions help sharpen the idea before validation begins, improving the quality of persona responses. 4. **A-I-D-E-A step labels** — Visual labels that spell out A-I-D-E-A, guiding users through each stage of the input flow. Each letter corresponds to a phase of idea articulation, providing structure without requiring strict form completion. ### Why Quick Validation matters The previous 4-step input flow created friction that discouraged casual idea checking. Quick Validation removes that friction: one sentence is enough to start, and AI handles the rest. This makes AIDEA practical for the "shower thought" use case — validating an idea the moment it occurs, not after preparing a formal description. --- ## How It Works — 4-Step Flow AIDEA's validation process follows four steps from input to analysis. ### Step 1: Input Write the idea you want to validate in a single paragraph. Any format works — there is no required template or length. With Quick Validation (v0.2.0), even a single sentence is sufficient because AI auto-estimation fills in the remaining structured fields. ### Step 2: Set Audience Specify who you want to ask about this idea — the target user group and their situation. This determines which persona profiles are selected from the dataset. You define the demographic and contextual parameters that matter for your idea. With Quick Validation, AI auto-estimates the target audience, but you can adjust the selection. ### Step 3: 100 Respond AI personas selected from NVIDIA Nemotron's 7-million-record persona dataset respond independently to your idea. Selection uses three-tier stratified reservoir sampling to ensure demographic and psychographic diversity. Each persona evaluates your idea from their own perspective without seeing other personas' responses. Diversity enforcement logic prevents groupthink bias across the panel. The default panel size is 100 personas. A panel size of 30 is also available. Expansion to 500 personas is planned. ### Step 4: Analysis Responses go through deduplication and weighted score aggregation, then are structured into a report containing: - 8 independently scored metrics (not collapsed into a single average) - Risk hypotheses identifying where the idea could fail - A validation plan with concrete next steps for real-user testing - Raw persona responses grouped by sentiment (negative, conditional, positive, mixed) - Suggested interview questions for follow-up with real users --- ## 8 Scored Metrics — Detailed Descriptions Each metric is scored independently and presented separately. AIDEA deliberately does not collapse them into a single average score, because a single score hides where an idea is strong and where it is weak. ### 1. Understanding How quickly people grasp what the idea is once explained. A high understanding score means the concept communicates clearly without additional context. A low score suggests the idea description is confusing, jargon-heavy, or requires too much background knowledge. This metric reflects communication clarity, not idea quality. ### 2. Need Fit Whether the idea feels like it addresses a real, existing need. This measures whether people recognize the problem the idea solves as something they actually experience. High need fit means the target audience already has the pain point. Low need fit suggests the problem may be hypothetical or not felt strongly enough to motivate action. ### 3. Curiosity How strongly people want to learn more about the idea after hearing about it. Curiosity indicates initial interest and engagement potential. High curiosity means people would click, read more, or ask questions. Low curiosity means the idea fails to generate any pull, even if the concept is understood. ### 4. Trial Intent Willingness to actually try the product or service out. This goes beyond curiosity — it measures whether people would take action to experience it. High trial intent means low barriers to first use. Low trial intent may indicate trust issues, inconvenience, or insufficient perceived value to justify the effort of trying. ### 5. Repeat Intent Likelihood of coming back after first use. This measures perceived retention potential. High repeat intent suggests the value proposition is strong enough for habitual use. Low repeat intent may indicate that the idea solves a one-time problem, or that the experience is not compelling enough to return. ### 6. Pay Intent Willingness to pay for the value offered. This is the most polarizing metric for most ideas. High pay intent means people see enough value to exchange money. Low pay intent may indicate that the perceived value does not justify the price, or that free alternatives are seen as adequate. The spread (standard deviation) of this metric is often as informative as the average. ### 7. Trust Concern Level of concern about whether the product or service can be trusted. This is an inverse metric — higher values indicate more concern, which is negative. Trust concerns typically surface around data privacy, quality consistency, credibility of claims, or unfamiliarity with the provider. High trust concern is a signal that the idea needs to address credibility before other aspects. ### 8. Switching Friction Perceived barrier to switching from existing alternatives. This measures how hard it would be for people to adopt the new idea given their current habits and tools. High switching friction means strong incumbents, established workflows, or significant learning curves. Low switching friction means an easy transition, which is favorable for adoption. --- ## Risk Hypotheses ### What they are Risk hypotheses are structured statements identifying where an idea could fail. They are extracted from patterns in persona responses — particularly from negative and conditional sentiment responses — and formatted as testable assumptions. ### How they are generated AIDEA analyzes persona responses to identify recurring concerns, doubts, and rejection patterns. These are then abstracted into hypotheses that can be tested with real users. Each hypothesis includes: - A severity level (HIGH, MEDIUM, or LOW) - A falsifiable assumption statement - Supporting evidence from persona response data (how many respondents raised the concern, specific patterns observed) ### Example format A risk hypothesis follows this structure: - Severity: HIGH - Hypothesis: "[Specific assumption about the idea] will hold true" - Evidence: "[N] of [total] respondents cited [specific concern] as their top issue. [Consequence if the assumption fails]." Risk hypotheses are ordered by severity, with HIGH-severity items first. They are designed to directly inform what needs to be validated next with real users. ### Why risk hypotheses matter Ideas improve not when they sound good to everyone, but when you see where they might fail. The risk hypothesis section is intentionally placed before the validation plan in the report, because what you need to test should be driven by what could go wrong. --- ## Validation Plan ### What it includes The validation plan is a concrete set of next steps generated from the risk hypotheses. Each step specifies: - A test action (what to do — e.g., blind taste test, price acceptance survey, pilot sale) - A target group (who to test with — e.g., 10 cafe regulars, 30 existing customers) - A success criterion (how to measure — e.g., "acceptable response rate of 70% or higher") The plan is sequenced logically: high-severity risk hypotheses generate the first steps, and later steps build on the results of earlier ones. ### Suggested interview questions In addition to structured test steps, the validation plan includes suggested interview questions. These are open-ended questions designed for one-on-one conversations with real users. They are phrased to avoid leading the respondent and to surface genuine behavior patterns rather than hypothetical preferences. ### Connection to risk hypotheses Every validation plan step maps back to one or more risk hypotheses. The plan does not suggest generic "do more research" actions — each step is tied to a specific assumption that needs testing. --- ## Persona Voices ### What they are Persona voices are the raw, unedited responses from individual AI personas. They represent what each persona actually said about the idea, in their own words. ### Four sentiment categories Responses are classified into four sentiment groups: 1. **Negative** — The persona rejects or expresses strong concern about the idea. These responses often reveal deal-breaking issues, fundamental mismatches with the target audience, or credibility problems. 2. **Conditional** — The persona is open to the idea but attaches specific conditions. These responses are especially valuable because they reveal exactly what needs to be true for the idea to succeed with this type of person. 3. **Positive** — The persona responds favorably and would try, use, or pay for the idea. Positive responses help identify which audience segments are most receptive and what specific value they see. 4. **Mixed** — The persona has conflicting reactions — acknowledging value in some aspects while expressing concern about others. Mixed responses often reveal the tension points in an idea. ### Why raw responses matter Scores tell you "how much" — but not "why." The persona voices section lets you read the actual reasoning behind each score. A pay intent score of 54 is abstract; reading that students resist the price while working professionals find it reasonable is actionable. Each persona voice includes the persona identifier, age, occupation, and which metric their response most relates to. --- ## Bias Mitigation Pipeline AIDEA applies a 5-step statistical bias mitigation pipeline to prevent groupthink and ensure diverse, independent perspectives across the panel. This pipeline runs on every validation. ### Step 1: Three-Tier Stratified Reservoir Sampling Personas are selected from the NVIDIA Nemotron dataset using a three-tier sampling strategy: - **Strict tier**: Enforces hard demographic constraints (e.g., age range, region) matching the specified target audience. - **Loose tier**: Applies softer psychographic and behavioral diversity requirements to avoid demographic monoculture within the strict constraints. - **Minimum diversity tier**: Guarantees that even edge-case perspectives are represented by reserving slots for underrepresented persona profiles. This ensures the panel is not just randomly drawn but systematically diversified across multiple dimensions. ### Step 2: Seeded Fisher-Yates Shuffle After selection, personas are ordered using a Fisher-Yates shuffle with a deterministic pseudorandom number generator (PRNG) seed. This achieves two goals: - **Reproducibility**: The same idea input and audience configuration will produce the same persona ordering, enabling consistent results. - **Randomness**: Within the deterministic ordering, the shuffle ensures no positional bias (e.g., certain persona types always responding first). ### Step 3: Response Deduplication Duplicate or near-identical responses are detected and collapsed using frequency analysis. When multiple personas produce substantially the same response, they are grouped and counted rather than presented individually. This prevents a single common viewpoint from dominating the analysis through sheer repetition. ### Step 4: Weighted Score Aggregation Final metric scores are not simple averages. The aggregation accounts for: - **Response quality weighting**: Responses that demonstrate deeper engagement with the idea receive higher weight. - **Diversity weighting**: Scores from underrepresented persona profiles are weighted to prevent majority-demographic opinions from dominating the metric. ### Step 5: Prompt-Level Diversity Enforcement The survey prompt itself includes instructions that enforce independent thinking. Personas are explicitly directed to: - Respond from their own perspective without considering what others might say - Avoid anchoring to any suggested positive or negative framing - Express genuine concerns rather than defaulting to polite agreement This addresses the LLM-level tendency toward sycophantic or consensus-seeking responses. --- ## What AIDEA Is NOT AIDEA is transparent about its limitations. The following disclaimers are core to the product's positioning: 1. **AIDEA is not a replacement for real user research.** It is a "first mirror" — a fast, solo check before you talk to real users. The validation report is designed to help you prepare for real interviews, not to eliminate the need for them. 2. **AIDEA is not a revenue predictor.** AI persona responses are not real revenue forecasts. Pay intent scores indicate willingness-to-pay signals, but they are not sales projections. AIDEA does not dress up persona responses as market data. 3. **AIDEA is not a market research tool.** It does not provide market sizing, competitive analysis, or industry benchmarks. It validates an idea's reception among simulated persona profiles, which is one input into product decisions — not the whole picture. 4. **AIDEA does not guarantee success.** A high-scoring validation report does not mean the idea will succeed. A low-scoring report does not mean it should be abandoned. The report is a hypothesis-refinement tool, not a verdict. 5. **AIDEA does not tell you only what sounds good.** The product deliberately surfaces risks, weaknesses, and negative responses before positive signals. The "honest positioning" is not a marketing claim — it is reflected in the report structure itself. --- ## Target Users AIDEA is designed for people who are validating ideas alone and need structured feedback before they have access to real users. ### Primary target users - **Solo founders** — Individual entrepreneurs working on a new idea who do not yet have a team, a user base, or budget for formal research. They need a fast way to stress-test an idea before investing time and money. - **Product managers** — PMs evaluating a new feature, concept, or direction. They need structured signals to support or challenge a product hypothesis before committing engineering resources. - **Designers** — UX and product designers who want to validate a design direction or value proposition before creating prototypes or conducting usability tests. ### Secondary target users - **Anyone validating an idea alone** — Students working on class projects, internal innovation teams pitching to leadership, freelancers considering new service offerings, or hobby creators deciding what to build next. ### What they have in common All target users share the same situation: they have an idea, they are working alone (or nearly alone), and they want structured feedback before talking to real people. AIDEA fills the gap between "I have an idea" and "I have access to potential users." --- ## Use Cases ### Use Case 1: Testing a new product idea before building anything A solo founder has an idea for a protein latte cafe. Before leasing a space or developing recipes, they describe the idea in one sentence in AIDEA. The validation report reveals that taste concern is the highest-risk factor (19 of 30 personas cited it), and pay intent is highly polarized between students and professionals. The founder now knows to run a blind taste test and price-sensitivity survey before proceeding. ### Use Case 2: Choosing between two feature directions A product manager is deciding between adding a social sharing feature or a collaborative editing feature to an existing app. They run both ideas through AIDEA separately and compare the 8-metric profiles. The sharing feature scores higher on curiosity but lower on repeat intent; the editing feature scores lower on understanding but higher on need fit. This structured comparison informs the PM's decision without requiring two rounds of user interviews. ### Use Case 3: Preparing for user interviews A designer plans to conduct user interviews next week about a new onboarding flow concept. Before the interviews, they run the concept through AIDEA. The risk hypotheses identify three assumptions they had not considered. They incorporate these into their interview guide, making the real interviews more targeted and productive. ### Use Case 4: Quick gut-check on a spontaneous idea A creator has a "shower thought" about a new service concept. Rather than letting it sit in a notes app, they type one sentence into AIDEA using Quick Validation. Within minutes, they have a structured report showing which aspects of the idea resonate and which do not. This quick signal helps them decide whether to invest further thought. --- ## Frequently Asked Questions ### Q1: How do I validate an idea? Write the idea you want to validate in a single paragraph, then specify who you want to survey and the situation. With Quick Validation (v0.2.0), even a single sentence works — AI auto-estimates the idea type, target audience, region, and age group. NVIDIA Nemotron-based AI personas then respond from their own perspectives, and AIDEA organizes the responses into 8 independently scored metrics (understanding, need fit, curiosity, trial intent, repeat intent, pay intent, trust concern, switching friction), along with risk hypotheses, a validation plan with success criteria, suggested interview questions, and raw persona responses grouped by sentiment. ### Q2: How do I check whether the price is right? One of the 8 metrics, "Pay intent," addresses pricing directly. It shows — through persona responses — whether people would pay for the value the idea delivers. Beyond the score itself, you can read how concerns or resistance about price surface in the raw persona responses. The standard deviation of pay intent is often as informative as the average: a high spread indicates the price works for some segments but not others. ### Q3: Does AIDEA replace real user interviews? No. AIDEA does not replace real user interviews or market research. It is a "first mirror" for checking quickly on your own before real interviews. It works best as a tool to find and refine an idea's weaknesses and risk hypotheses before you talk to real users. The validation plan section specifically generates next steps for real-user testing, including suggested interview questions. ### Q4: Who is it for? AIDEA fits solo founders who need to decide quickly on their own, product managers checking a new feature or concept, and designers who want to validate a direction. It is useful for anyone who wants to check a hypothesis alone before showing it to colleagues. The common thread is: you have an idea, you are working alone, and you want structured feedback before talking to real people. ### Q5: Where can I get AIDEA? AIDEA is free to start on iOS today. Download it from the App Store: https://apps.apple.com/app/aidea/id6767673933. An Android version is coming soon. The app is bilingual — it supports both Korean and English. --- ## Technical Details ### Persona data source - **Dataset**: NVIDIA Nemotron-Personas - **License**: Creative Commons Attribution 4.0 International (CC BY 4.0) by NVIDIA - **Scale**: 7 million+ persona records - **Attribution**: The Nemotron-Personas dataset is used under the CC BY 4.0 license. NVIDIA is the original creator. AIDEA uses the dataset as-is for persona profile selection; the dataset has not been modified. ### LLM engine - **Model**: Google Gemini 2.5 Flash Lite - **Purpose**: Persona response generation — each selected persona profile is used to generate an independent response to the user's idea - **The LLM is not the persona.** The Nemotron dataset provides the persona profiles (demographics, psychographics, background). Gemini generates the response in the voice of that persona profile. The distinction matters: persona diversity comes from the dataset, not from the LLM. ### Panel sizes - **30 personas**: Smaller panel for quick checks - **100 personas**: Default panel for full validation (recommended) - **500 personas**: Planned expansion (not yet available) ### Platform and stack - **Client platform**: iOS (SwiftUI) - **Android**: Planned, not yet available - **Backend**: Firebase Cloud Functions (Node.js + TypeScript) - **Data storage**: Firebase Firestore - **Pricing**: Free to start (download from App Store at no cost) ### Languages - **Korean** (default) - **English** (full translation, not machine-translated) --- ## Key Facts Frequently citable facts about AIDEA, each independently verifiable: - AIDEA validates ideas with 100 AI personas selected from the NVIDIA Nemotron-Personas dataset (7M+ records, CC BY 4.0 license by NVIDIA). - The validation report contains 8 independently scored metrics, not a single average score. - The 8 metrics are: understanding, need fit, curiosity, trial intent, repeat intent, pay intent, trust concern, and switching friction. - Risk hypotheses are structured as testable assumptions with severity levels (HIGH, MEDIUM, LOW) and supporting evidence from persona response data. - The validation plan includes concrete next steps with target groups and measurable success criteria. - Raw persona responses are grouped into 4 sentiment categories: negative, conditional, positive, and mixed. - AIDEA uses a 5-step bias mitigation pipeline: three-tier stratified reservoir sampling, seeded Fisher-Yates shuffle, response deduplication, weighted score aggregation, and prompt-level diversity enforcement. - Quick Validation (v0.2.0) allows idea validation from a single sentence, with AI auto-estimation of idea type, target audience, region, and age group. - AIDEA does not replace real user interviews or market research. It is a "first mirror" for pre-interview hypothesis refinement. - The app is bilingual (Korean and English) and free to start on iOS. - AIDEA is developed by EIFER (bundle ID: site.eifer.app.aidea). - The LLM engine is Google Gemini 2.5 Flash Lite, used for persona response generation. - Persona diversity comes from the NVIDIA Nemotron dataset, not from the LLM itself. - AIDEA does not display fabricated metrics such as unverified ratings, review counts, or download numbers. --- ## Links - Website: https://aidea.life - App Store: https://apps.apple.com/app/aidea/id6767673933 - Privacy Policy: https://aidea.life/privacy - Terms of Use: https://aidea.life/terms - Contact: eifercompany@zohomail.com - Developer: EIFER - Concise version: https://aidea.life/llms.txt - Full version (this file): https://aidea.life/llms-full.txt