Case study · Market Research / AI

Sayso: AI-assisted market research with fraud-filtered panel data

Client
Sayso
Industry
Market Research / AI
Location
Global
Services
AI Solutions, Product Engineering, Backend Development, DevOps Consulting

The client

Sayso is a consumer survey and market-research platform. Panelists register, complete a demographic profile, take paid surveys and earn rewards, while researchers create surveys and measure participation and response quality.

The brief: use AI to draft survey questions from uploaded source documents, keep a human approval step before anything goes live, and score every answer so fraudulent respondents never reach the data buyers receive. A companion mobile app and admin analytics dashboards complete the product.

0–100
Response-quality score
3
Isolated cloud environments

01 · The challenge

What stood in the way

Response fraud

Survey panels suffer from speedsters and low-effort respondents. Data buyers needed confidence that every response in a dataset is genuine.

AI with human oversight

Generating survey questions from source documents with AI is fast, but every draft needed human approval before reaching panelists.

Reaching the right panelists

Surveys only produce good data when they reach the right respondents, so distribution had to be driven by profile matching, not blasts.

Clean multi-environment delivery

Development, staging and production had to stay cleanly separated in three isolated cloud environments with branch-per-environment promotion.

02 · The solution

How we built it

  1. Response-quality engine

    Every response is scored 0–100, detecting speedsters and sub-two-second answers. Suspect respondents are flagged so buyers receive clean data.

  2. AI question pipeline

    Admins upload a source PDF; the system extracts context and drafts survey questions for human review before distribution, tracking model confidence per question.

  3. Profile-scored targeting

    A profile-scoring engine ranks panelists for each survey before distribution, so it reaches the respondents most likely to answer well.

  4. Complete survey lifecycle

    Categories, surveys and questions with conditional logic, resumable survey-taking, multi-typed answer storage and a configurable reward engine.

  5. Production-grade backend

    Strict TypeScript, a layered architecture, typed error envelopes, Zod validation and Aurora reader/writer pool splitting for scale.

  6. Automated cloud delivery

    Docker multi-stage builds are promoted through three isolated AWS environments by branch-mapped GitHub Actions pipelines.

03 · The results

What changed for the client

Cleaner panel data

The quality engine filters out speedsters and flags suspect respondents, protecting the value of every dataset sold.

Surveys in minutes

AI-drafted questions with human approval cut survey creation from hours to minutes without lowering quality.

Precise targeting

Profile-scored distribution sends each survey to the panelists most likely to give quality answers.

Stable, auditable delivery

Three isolated environments and automated pipelines give the research business a stable, auditable platform.

Technology used

  • Node.js 20
  • TypeScript
  • Express 5
  • Zod
  • PostgreSQL (AWS Aurora)
  • Drizzle ORM
  • LLM question generation
  • Vector embeddings
  • AWS ECS
  • AWS ECR
  • Docker
  • GitHub Actions
  • CloudWatch

Planning something like Sayso?

Tell us about your project. A senior engineer will reply within one business day with questions, a suggested approach and next steps.