Mansi
HOMEWORKASKABOUTCONTACT
HOMEWORKASKABOUTCONTACT
Resume

Home/Work

Project 02

Productionizing an ML-Driven Allocation Engine

Signal to verdict — predictions become allocation decisions.

Lead Data Engineer · Data Platforms · Cloud Architecture · ETL ModernizationMachine Learning · Data Engineering · AWS

Amazon S3DatabricksDeltaMLflowAmazon ECREC2
Allocation engineSignals become eligibility, classification, allocation, then a user-facing recommendation.SignalsDecisioningServingInfrastructureUser signalsEligibilityClassificationpersonasAllocation logicGraphQLrecommendationS3 / DeltaDatabricksMLflowECR + EC2
Signals become eligibility, classification, allocation, then a user-facing recommendation.

01 — The problem

Users signed up for paid tiers but did not consistently adopt the broader product stack; the team needed a decisioning layer—not another funnel—to drive relevant allocations.

02 — The system

Select a layer to see its role.

Technology — why it was used

Amazon S3

Databricks

Delta

MLflow

Amazon ECR

EC2

GraphQL

05 — Evidence

  1. 01DataS3 · Databricks
  2. 02Model development
  3. 03Model trackingMLflow
  4. 04ContainerizationAmazon ECR
  5. 05DeploymentEC2
  6. 06ServingGraphQL
The model was one stage. Production had to carry the decision to the user.

Takeaway

A model creates predictions. A production system turns them into decisions.

What landed

  • Delivered end-to-end path from model development through serving
  • Separated classification, allocation logic, processing, and API layers
PreviousLegacy Data ModernizationAll workNextWeb Intelligence Pipeline
Mansi

© 2026 Mansi Dhruv. All rights reserved.

Created with curiosity