Embedded Files
Personalized Recommendations for Walmart Business
Personalized Recommendations for Walmart Business
Integrated personalized recommendation modules into Walmart Business Cart and Post Add to Cart (PAC) experiences across desktop, mobile web, iOS, and Android.
Impact at a Glance
Impact at a Glance
2 Pages
2 Pages
Customer Journey Coverage
Integrated personalized recommendations into both Cart and Post Add to Cart (PAC) experiences.
4–5 Modules Per Page
4–5 Modules Per Page
Recommendation Coverage
Enabled approximately 4–5 recommendation modules on each page.
4 Platforms
4 Platforms
Cross-Platform Reach
Supported the shared experience across desktop, mobile web, iOS, and Android.
1 Recognition
1 Recognition
Engineering Contribution
Received an internal Walmart recognition for contributions to the personalization initiative.
Problem / Engineering Context
Problem / Engineering Context
Navigating a Complex Shared Platform
Navigating a Complex Shared Platform
Walmart Business wanted to replace static recommendations on Cart and PAC with personalized recommendation modules already supported by Walmart's broader personalization ecosystem.
The challenge was that the functionality lived inside a large shared monorepo used by multiple business units. Documentation was incomplete, internal model names were cryptic, and active execution paths were mixed with obsolete configuration and legacy logic.
There was no formal knowledge transfer for the existing B2C implementation, so I had to independently trace the system, identify the active paths, validate my findings with B2C points of contact, and determine how Walmart Business should integrate safely.
How the Integration Worked
How the Integration Worked
1. Trace Existing Paths
1. Trace Existing Paths
Located the existing B2C Cart and PAC execution paths inside the shared personalization monorepo.
2. Verify Active Logic
2. Verify Active Logic
Manually traced model/module selection, fallbacks, and configuration to distinguish active behavior from obsolete code.
3. Register B2B Tenant
3. Register B2B Tenant
Configured and registered the Walmart Business tenant within the shared personalization service.
4. Enable Recommendation Modules
4. Enable Recommendation Modules
Integrated approximately 4–5 recommendation modules on each of Cart and PAC.
5. Roll Out & Observe
5. Roll Out & Observe
Capabilities were progressively exposed through Walmart's internal Expo experimentation/percentage-rollout platform while latency and errors were monitored through openObserve.
What I Personally Owned
What I Personally Owned
Reverse-engineered the existing B2C Cart and PAC personalization paths inside a large, poorly documented shared monorepo.
Used GitHub Copilot and Cursor as code-navigation/search aids, while manually validating the actual execution paths.
Partnered with B2C points of contact to verify active configuration and model-to-module behavior.
Configured and registered the Walmart Business tenant integration.
Integrated approximately 4–5 recommendation modules on each of Cart and PAC.
Built openObserve dashboards for latency and error monitoring.
Created onboarding documentation covering code paths, terminology, model-to-module mappings, and active versus obsolete configurations.
Contributed backend telemetry interfaces and schemas supporting user-journey events.
Supported the phased production rollout across desktop, mobile web, iOS, and Android.
Engineering Decisions I Drove
Engineering Decisions I Drove
Verify Before Integrating
Verify Before Integrating
Instead of trusting undocumented configuration or AI-assisted search results, I manually traced the relevant code paths and validated my understanding with existing B2C owners before changing the B2B integration.
Isolate Active Configuration
Isolate Active Configuration
I separated active personalization logic from obsolete and legacy configuration so the B2B integration could avoid depending on dead paths.
Build Documentation While Reverse-Engineering
Build Documentation While Reverse-Engineering
I converted what I learned into onboarding documentation covering terminology, execution paths, model/module mappings, and active configuration so future engineers would not have to repeat the same discovery work.
Recommendation Flow
Recommendation Flow
I understood the surrounding selection and fallback behavior sufficiently to integrate the B2B tenant correctly, but I did not build or modify the underlying Arbitration Engine.
Production Rollout & Observability
Production Rollout & Observability
Phased Rollout
Phased Rollout
Recommendations were initially exposed to a smaller customer population and progressively increased through Expo, Walmart's internal experimentation and percentage-rollout platform.
Observability
Observability
Built openObserve latency and error dashboards to monitor Cart and PAC recommendation integrations in production.
Knowledge Transfer
Knowledge Transfer
Created documentation that was later used to help onboard engineers joining the personalization work.
Collaboration & Ownership Boundaries
Collaboration & Ownership Boundaries
Arbitration Engine
Arbitration Engine
Not owned by me
The underlying Arbitration Engine selected model outputs for recommendation slots. I investigated and understood its behavior for integration purposes but did not build or modify it.
Recommendation Models
Recommendation Models
Not owned by me
I did not build, train, or own the recommendation models themselves.
Expo Platform
Expo Platform
Not owned by me
Expo was Walmart's existing experimentation and rollout platform. I worked within the rollout process but did not build the platform.
Full-Market Launch
Full-Market Launch
Not owned by me
The functionality was progressively rolled out, but I left Walmart before exposure reached 100%.
AI-Assisted Development
AI-Assisted Development
Used as navigation support
GitHub Copilot and Cursor helped me search and navigate the large repository. I retained responsibility for manually tracing, verifying, and implementing the integration.
My core ownership: B2B tenant integration, Cart/PAC code-path discovery and verification, recommendation-module enablement, production observability, telemetry interfaces, and onboarding documentation.
Bottom CTA
Bottom CTA
Page updated
Google Sites
Report abuse