Skip to content
Sharif Amlani, PhD

Twenty-five hours of survey analysis.
Ten minutes.

As CTO of HumanAI Ventures I am building Survey Fluency, an AI platform that turns raw survey data into toplines, cross-tabs, executive summaries and a PowerPoint deck.

surveyfluency.com

Role
Chief Technology Officer (CTO)
Organization
HumanAI Ventures · Survey Fluency
Years
2025 – Present · Sacramento
Stack
LangChain · LangGraph · Claude + OpenAI APIs · FastAPI · Docker · GCP · Supabase

The outcome

Same analysis.
150 times faster.

25 hoursManual analyst workflow · 1,500 minutes

Drawn to scale. Twenty-five hours is 1,500 minutes; ten minutes is 1/150 of the bar.

10 min
Per analysis
Down from 25 hours
95%
Accuracy
Agent-run statistical analyses
483
Surveys analyzed
For 147 researchers · Oct 2026
3
Design-partner pilots
KL Communications · Beacon Research · C+R Research

01 — The problem

Survey analysis ran
on analyst hours.

Turning raw survey data into toplines, cross-tabs, an executive summary and a client-ready PowerPoint deck is manual analyst work: a 25-hour workflow for political pollsters and market research firms.

02 — What I built

Agents that plan,
run and critique.

  • Designed and deployed LangChain and LangGraph pipelines on the Claude and OpenAI APIs that autonomously plan, execute and critique statistical analyses, at 95% accuracy.
  • Upload a CSV, Excel, SPSS or Stata export and get cross-tabs with significance testing, open-end coding, regressions and a client-ready report.
  • Built and deployed the backend API on Google Cloud with Docker, CI/CD, FastAPI and Supabase authentication.
  • Running discovery with political pollsters and market research firms, and active design-partner pilots with KL Communications, Beacon Research and C+R Research.
  • Alongside it, Synthetic Respondents: AI digital twins that predict survey response distributions across demographic subgroups.
Stage 01

Plan

Plans the statistical analysis the survey calls for.

Stage 02

Execute

Runs it: toplines, cross-tabs with significance testing, open-end coding and regressions.

Stage 03

Critique

Critiques its own analysis before the summary and the deck are written.

03 — How it works

Raw data in.
A deck out.

Fig. 01 — How the pipeline fits together
Simplified
  1. 01 · UploadA raw survey export goes in: CSV, Excel, SPSS or Stata.
  2. 02 · PlanThe agents plan the statistical analysis.
  3. 03 · ExecuteThey run it: toplines, cross-tabs with significance testing, open-end coding, regressions.
  4. 04 · CritiqueThey critique the analysis, then write the executive summary and the deck.

04 — How we knew it worked

Measured,
not guessed.

An agent that is fast and wrong is a liability. These are the results on the record.

Speed25 hrs → 10 minA 25-hour manual analyst workflow, from raw survey data to a client-ready report, in about 10 minutes.
Accuracy95%Of the statistical analyses the agents plan, execute and critique.
Usage483 surveysAnalyzed for 147 researchers, as of October 2026.
Pilots3 design partnersKL Communications, Beacon Research and C+R Research.

05 — In their words

In pilots with design partners: KL Communications, Beacon Research, C+R Research.

HumanAI Ventures · Active design-partner pilots

Founder in UC Berkeley SkyDeck's Pad-13 cohort, 2026: about 3% acceptance.