L'Atelier BNP Paribas
/Client
Spotting change early
L'Atelier's models scan technology and social data and flag unusual changes for researchers to investigate. The bank's own researchers now run the system.
Custom algorithms and AI products.
For over a decade, DataSciences has built custom models and put them into working software at more than 45 companies.
A particle filter tracking a noisy signal, live
Selected clients
BNP Paribas
Paysafe- L'Atelier BNP Paribas
Airtm
Mursion
Anthemis
Cognisity
HouseQuest- CFCamino Financial
ForwardLine- AAspiration
Ikigai Data
Admissions GPS- UAUpdate AI
HundredX
Fintel Connect
Massumi + Consoli
Open Sesame Media- SSocialeads
Burst Oral Care
ChatterWorks
MagicLinks- OWOro Wellness
Userwave- JJigsaw
Lifeed- RSRecord Store Studio
- DLDivining Labs
45+
companies served
Lenders and payments companies, a global bank, venture investors, health and media apps, property tax and real estate platforms, education, and law.
$100M+
in loans decided by the credit models
Consumer and small business credit models for lenders.
500K
scooters across Bird's cities
Placed each day by the allocation engine to get the most rides from the fleet.
20+
models and AI systems in production
Credit, fraud, allocation, property assessment, recommendation, trend detection, and conversational AI.
10+
companies advised on AI and data
Board and advisory roles in lending, recruiting, real estate, careers, and nonprofits.
65+
people led across AI and data teams
Company teams, client R&D groups, and DataSciences project teams.
2
granted US patents in psychometrics
One on credit risk from psychometric answers, one on personality from social media language.
8,300+
academic citations
Including a landmark 2004 PNAS paper that mapped how the human brain develops.
Case studies
The animations use generated data, no client data.
01 / 13
A
AI Responsibility Lab
/Client
Making generative AI safe at work
A safety layer for generative AI at work scores every prompt for risk before it reaches the model. Personal data is masked, and doubtful requests are held for review.
S
Socialeads
/Client
Detecting life events in language
Language models read public posts as they are written and pick out life events and buying intent, such as a new job or a move. The posts the model is least sure of go to people to label, so it keeps learning. The system ran in production for three years.
D
Divining Labs
/Client
Validating flood risk models
Predicted flood depths were checked building by building against measured depths. Each error was converted into dollars of loss, which shows where a wrong prediction costs the most.
Massumi + Consoli
/Client
Bringing AI into a deal practice
We measured the firm's deal work for time and risk, then tested the leading legal AI tools against it. The firm adopted a phased plan.
Nimble.LA Group
/Client
Pulling claims out of company reports
Language models read company websites and reports and pull out every sustainability claim. Each claim is scored for quality before it is kept, then grouped by theme with repeats merged, so analysts compare companies claim by claim.
Fintel Connect
/Client
Verifying advertising content
Large language models check advertising content against the rules it has to follow. Guardrails tie every finding to the passage behind it, which keeps the checks precise at volume.
Mursion
/Client, then embedded role
AI for difficult conversations
Mursion runs live simulations for practicing difficult conversations at work. The AI work covered speech-to-speech conversation, automated scoring, and coaching feedback from transcripts.
Cognisity
/Client
AI-driven property assessment
Deep LLM reasoning models work through each property in small steps, and guardrails and quality checks test every step before the next one runs. Properties whose assessed value falls outside the estimate are set aside for review, with the evidence for each.
HouseQuest
/Client
Matching people with homes
HouseQuest measures what each buyer cares about across many dimensions of preference. Homes are then ranked by fit, which search filters cannot capture.
R
Record Store Studio
/Client
Building person‑level ontologies
Large image models and intelligent decision systems organize a photo archive around the people in it. Each person becomes one connected record, linked to every photo and story they appear in.
Bird
/Embedded role
Putting scooters where riders are
Bird's allocation engine decided where hundreds of thousands of scooters were placed each day. The team also built Bird's fraud and loss models and its analytics for city governments.
P
Payoff
/Embedded role
Psychometrics for lending
Payoff's data science team built the Financial Personality Quiz and models for credit risk, fraud, and identity. Fraud held at zero for two years, and the credit method became a US patent.
Services
New models and AI features, plus testing and rebuilding of systems already in use.
Prediction, scoring, and pricing
Models for credit risk, fraud, demand, property assessment, and pricing, running in production at lenders and platforms.
Recommendation and matching
Recommendation systems for home buyers, students, and job seekers that rank choices by fit.
Language, vision, and detection
Systems that analyze documents, survey answers, social posts, images, and interviews. Past work includes trend detection, face clustering, and conversational AI.
Optimization, validation, and rebuilds
Allocation systems, validation of existing models, and new scoring systems to replace old ones, such as the rebuilt loan and renewal scorecards at ForwardLine.
Team
DataSciences is led by David Herman, PhD, a co-inventor on two granted US patents. He has led AI and data teams inside client companies, as an executive, and on DataSciences projects with senior specialists from his network.
Teams led
DataSciences
Client project teams
Cognisity
Algorithm R&D team
HouseQuest
AI R&D team
Kanope
Product and engineering team
L'Atelier BNP Paribas
Trend research team
AI Responsibility Lab
Machine learning team
Socialeads
Language model team
Mursion
AI and data science team
Bird
Data organization of 50+
Payoff
Data science team of 10
psyML
Founding team
Brought into DataSciences projects
- Machine learning engineers
- Data scientists
- Data and software engineers
- AI researchers
- Psychologists and psychometricians
- Product designers
- Industry advisors
- Nimble.LA Group, partner firm
Senior specialists from his network join each engagement as the work requires. Several now hold senior and VP roles at other companies.
Experience
Current
D
DataSciences
Founder
More than 45 companies. Built credit models for Paysafe, ForwardLine, and Camino Financial, trend detection for L'Atelier BNP Paribas, and marketplace and venture analytics for Airtm and Anthemis. Leads algorithm R&D for Cognisity and HouseQuest.
Scorecards, recommendation, trend detection, psychometrics, language models
2024 to 2025
Mursion
VP, AI and Data Science
A DataSciences client from 2024, then VP in 2025. Led the AI and data science team behind Mursion's AI actor, which holds live speech-to-speech practice conversations scored from their transcripts.
Speech to text, large language models, text to speech, transcript scoring
2018 to 2019
Bird
Director of Data Science, then Director of Data Analytics and Science
Built the data science and analytics organization to more than 50 people in under a year. Its allocation engine set where scooters were dropped each day, and the team built fraud and loss models and analytics for city governments.
- people in the data organization
- 50+
- people in the data organization
Demand forecasting, fleet allocation, champion and challenger testing, anomaly models
2017 to 2019
P
psyML
Co-founder and CEO
2014 to 2017
P
Payoff
Data Scientist to VP of Data Science
Built a data science team of 10 and the company's models for credit risk, fraud, identity, and psychometrics, including the Financial Personality Quiz. Presented to investors in the Series B and C rounds. The psychometric credit method became a granted US patent.
- investor meetings, Series B and C
- 40+
- investor meetings, Series B and C
Credit scoring, fraud detection, psychometric assessment
2003 to 2014
A
Academic research
Neuroscience at NIMH, Baylor, and USC
Mapped human brain development at the National Institute of Mental Health, studied the amygdala at Baylor, and earned a PhD in computational neuroscience at USC, building a robotic system with machine vision to measure behavior. Eleven peer-reviewed papers, cited more than 8,300 times.
MRI brain mapping, electrophysiology, robotics, machine vision
Clients and advisory roles
Admissions GPS
Client
College fit scoring for students.
AI Responsibility Lab
Client
Trust scoring, filtering, and moderation for generative AI products.
Airtm
Client
Marketplace matching, fee, and pricing analysis for a payments platform.
Anthemis
Client
A data driven process for finding and assessing ventures.
Aspiration
Client
Credit modeling and data engineering.
AudioCardio
Client
Engagement and outcome analysis for a hearing health app.
Bambino
Client
A ranking model that matches families with sitters.
BrightQuery
Client
Credit risk analysis for a company data platform.
Burst Oral Care
Client
Statistical analysis of a toothbrush clinical study.
ByteSized Capital
Client
Technical due diligence on a startup before investment.
Camino Financial
Client and advisor
Credit score and prequalification models for small business loans.
ChatterWorks
Client
Built the candidate sourcing and engagement platform.
Cognisity
Client
Property tax assessment algorithms.
Divining Labs
Client
Statistical validation of flood depth models.
Fintel Connect
Client
Language model verification of affiliate program details.
ForwardLine
Client
Rebuilt the initial and renewal risk scorecards for small business loans.
Foster Nation
Board director
Board work on technology, AI, and fundraising.
Hangtight
AI advisor
AI strategy for the founding team.
HouseQuest
Client
Psychometric home matching and the product around it.
HundredX
Client
Customer personas and language processing from survey answers.
Ikigai Data
Advisor and partner
Data strategy, psychometrics, and career recommendation models.
Jigsaw
Client
Founder scoring models for a venture firm.
Kanope
Client
Product development for production management software.
L'Atelier BNP Paribas
Client
Technology and social trend detection, then a foresight system handed to the bank's researchers.
Lifeed
Client
AI strategy for a life skills platform.
MagicLinks
Client
Data science hiring and candidate interviews.
Massumi + Consoli
Client
AI strategy and vendor selection for an M&A law firm.
Milieu Technologies
Advisor
Strategy advisor to the founders.
Nimble.LA Group
Client and advisor
Sustainability claim extraction from company documents.
OC4 Venture Studio
Coach
Industry coach for portfolio companies.
Open Sesame Media
Client
Machine learning for a karaoke app ahead of its beta.
Oro Wellness
Client
A product recommendation engine and consumer survey for a wellness brand.
Paysafe
Client
A credit model for point of sale loans, built on credit bureau data.
Record Store Studio
Client
Computer vision and search for a personal archive product.
Sentio Solutions
Board observer
Board observer for a venture investor.
Socialeads
Client
Language models that read social posts for life events and buying intent.
Update AI
Client
Finding action items and decisions in meeting transcripts.
Userwave
Client
Language models for review summaries and question answering.
New projects
Send new project inquiries by email.