Swarmgram · AI Research
Behavioral AI for
research and simulation.
Swarmgram builds Lewis — a behavioral model stack calibrated against 460 real polling questions — and Lewsearch, an AI-native synthetic research panel that answers in minutes, not weeks, at a price point the teams who need it most can actually afford.
Looking for Lewsearch? Run studies here.
Lewsearch is the live product. Run synthetic audience studies, Creative Reviews, and U.S. national or regional reads at lewsearch.com.
Run studies →What we build
Company. Model. Product.
Company
Swarmgram
AI research company building behavioral models and synthetic population infrastructure. Delaware C-corp, headquartered in Columbus, OH.
Model
Lewis
A proprietary behavioral model stack powering 170,000+ demographically anchored agents across live city and state markets. Agents are refined against real-world data over time — the panel sharpens with use. Calibrated against 460 ground-truth polling questions.
Product
LIVELewsearch
AI-native synthetic research panels built for the teams traditional providers priced out — SMBs, agencies, and growth teams. Run studies, Creative Reviews, and ACS-weighted U.S. national or regional reads. First 500-agent study free. Paid studies from $19/credit. ±4.88% Texas MAE across 460 benchmarked questions.
Run a free study →Validation
Benchmarked against real-world polling data.
460
GT-anchored questions
Texas (120) · California PPIC (175) · 151-Q multi-metro · 22-Q held-out panel
±4.88%
Texas MAE · 5-fold CV
ex-electoral · calibrated · UT/TPP ground truth · full methodology public
170K+
Live agents
Live city/state markets · ACS-weighted · national and regional directional audiences
Free
500-agent study
Work-email signup credit · full PDF report · paid from $19/credit
Contact
Talk to the team.
Request a walkthrough of Lewsearch, benchmark methodology, or investment materials.
Or email directly: hi@swarmgram.com
Founder
Jeffrey Mead
3× founder · MBA · background in law enforcement, CPG operations, sales, and finance. Built Lewis end-to-end — model training, inference infrastructure, and both product surfaces in-house. I was the buyer. I know the pain firsthand.