04Experience
University of Toronto Machine Intelligence Student Team (UTMIST), in collaboration with Flybits
Machine Learning Engineer
On a six-person UTMIST team working with Flybits, helped build a privacy-preserving prototype for personalized digital-credit offers.
- Python
- Pandas
- NumPy
- scikit-learn
- Clustering
- Autoencoders
- Synthetic data
- Model evaluation
- InputSynthetic persona inputs
- PrepareValidation + preprocessing
- RepresentRepresentation learning
- ClusterClustering
- InterpretArchetype interpretation
Public-scope workflow+2 more
Contributions
- Processed and analyzed more than 100,000 synthetic customer personas.
- Explored clustering and autoencoder approaches for representation and segmentation.
- Helped identify more than five interpretable archetypes.
- Contributed to a six-person machine-learning engineering team.
- Supported the prototype's analysis, evaluation, and communication.
Context
A UTMIST student-team collaboration with Flybits. The inputs were synthetic personas, not real bank or customer records, and the work was a prototype rather than a production financial product or deployed credit-decision system.
In one line
Developed privacy-preserving customer archetypes from more than 100,000 synthetic personas for a personalized digital-credit-offer prototype.
Measured outcomes
100,000+
Synthetic customer personas processed
Scale of the generated persona dataset as reported in the current resume and LinkedIn record.
Synthetic personas — no real bank or customer records.
5+
Interpretable customer archetypes identified
Archetype count reported for the prototype segmentation work.
6
Person machine-learning engineering team
Team size as reported in the current LinkedIn record.
Tools & stack
- Python
- Pandas
- NumPy
- scikit-learn
- Clustering
- Autoencoders
- Synthetic data
- Model evaluation
Scope of this page
No public repository or live demo exists for this work, and none is linked. Flybits branding and any partner-internal material are deliberately absent.
- Synthetic personas only; no real-customer experimentation.
- No claim of production deployment, credit approval, underwriting, fairness, revenue, conversion, or model-lift outcomes.