Tags
Browse posts by tag
ai_agents
The Score Drift Agent
Jan 30, 2026 | by Aditya K., Matt F., Matt M.
Background In the past year, Branch has continued to experiment with ways that generative AI can be used in the practice of making loans. We call this Generative Credit.
attention
From Rules to Representation: Turning Event Sequences into User Vectors
Jul 03, 2026 | by Nishant Kumar, Vigneshmurali Iyer, Navin Kumar M.
Can a compact learned vector replace a wide surface of hand-engineered event-stream features without losing signal? We trained a small neural encoder on user financial event seq...
change_data_capture
From 12 Hours to 10 Minutes: Rebuilding Data Platform with CDC
Aug 25, 2026 | by Abhinav Kumar
We redesigned our data warehouse ingestion with an event-driven CDC pipeline, cutting warehouse data lag from 12 hours to 10 minutes.
credit_scoring
From Rules to Representation: Turning Event Sequences into User Vectors
Jul 03, 2026 | by Nishant Kumar, Vigneshmurali Iyer, Navin Kumar M.
Can a compact learned vector replace a wide surface of hand-engineered event-stream features without losing signal? We trained a small neural encoder on user financial event seq...
The Score Drift Agent
Jan 30, 2026 | by Aditya K., Matt F., Matt M.
Background In the past year, Branch has continued to experiment with ways that generative AI can be used in the practice of making loans. We call this Generative Credit.
distributed_systems
From 72 Hours to 8: Rebuilding our Feature Fetch System
Apr 30, 2026 | by Navin Kumar M., Aviral Sharma
How we engineered a parallel execution system to fetch features for training, achieving an 8× speedup through distributed actor-based design.
feature_engineering
From Rules to Representation: Turning Event Sequences into User Vectors
Jul 03, 2026 | by Nishant Kumar, Vigneshmurali Iyer, Navin Kumar M.
Can a compact learned vector replace a wide surface of hand-engineered event-stream features without losing signal? We trained a small neural encoder on user financial event seq...
feature_pipelines
From 72 Hours to 8: Rebuilding our Feature Fetch System
Apr 30, 2026 | by Navin Kumar M., Aviral Sharma
How we engineered a parallel execution system to fetch features for training, achieving an 8× speedup through distributed actor-based design.
kafka
From 12 Hours to 10 Minutes: Rebuilding Data Platform with CDC
Aug 25, 2026 | by Abhinav Kumar
We redesigned our data warehouse ingestion with an event-driven CDC pipeline, cutting warehouse data lag from 12 hours to 10 minutes.
postgres
From 12 Hours to 10 Minutes: Rebuilding Data Platform with CDC
Aug 25, 2026 | by Abhinav Kumar
We redesigned our data warehouse ingestion with an event-driven CDC pipeline, cutting warehouse data lag from 12 hours to 10 minutes.
ray
From 72 Hours to 8: Rebuilding our Feature Fetch System
Apr 30, 2026 | by Navin Kumar M., Aviral Sharma
How we engineered a parallel execution system to fetch features for training, achieving an 8× speedup through distributed actor-based design.
sequence_modeling
From Rules to Representation: Turning Event Sequences into User Vectors
Jul 03, 2026 | by Nishant Kumar, Vigneshmurali Iyer, Navin Kumar M.
Can a compact learned vector replace a wide surface of hand-engineered event-stream features without losing signal? We trained a small neural encoder on user financial event seq...
user_embeddings
From Rules to Representation: Turning Event Sequences into User Vectors
Jul 03, 2026 | by Nishant Kumar, Vigneshmurali Iyer, Navin Kumar M.
Can a compact learned vector replace a wide surface of hand-engineered event-stream features without losing signal? We trained a small neural encoder on user financial event seq...
xgboost
From Rules to Representation: Turning Event Sequences into User Vectors
Jul 03, 2026 | by Nishant Kumar, Vigneshmurali Iyer, Navin Kumar M.
Can a compact learned vector replace a wide surface of hand-engineered event-stream features without losing signal? We trained a small neural encoder on user financial event seq...