IBU PODx | March 2026 — Pilot Edition

From the Editor

A Pilot Worth Running

What we learned shaping the format before launch

March was the month we stopped sketching and started shipping. The opti_processor pipeline finally cleared its last channel — events processing for Canada — meaning we now have a unified processor for every market and every channel type. That's the spine of the NBA operation.

This pilot edition exists to test one thing: can we tell the team's story month over month without losing the data backbone? Answer so far: yes, but it needs tighter section curation. April will tell us if we've got the rhythm right.

Vaibhav Rachalwar · IBU Omnichannel / ML · Eli Lilly, Bengaluru

Numbers

March 2026 KPI Snapshot

The metrics that mattered this month

7

Channels Processed

VAE · Events · HQ · IVA · Calls · LillyPlay · LillyApp

518M

Rows Streamed

↑ IVA pipeline · pyarrow batching

5

Markets Live

EU M5 · META · LATAM · Japan · Canada

Project Spotlight

opti_processor — Now Fully Unified

One processor. Seven channels. Every market.

◆ Pipeline · Production Ready

Seven Channel Types, One Codebase

After resolving the IVA kernel crashes (the 518M-row file that broke SageMaker) via streaming 1M-row batches with pyarrow pre-filtering, every channel now flows through a single unified PipelineConfig dataclass. HQ product format mismatch resolved. Canada-specific BRD defaults wired in. The unified processor now serves EU M5, META, LATAM, Japan, and Canada with zero hardcoding.

PipelineConfig s3_data_manager opti_processor.py pyarrow streaming 5 markets

People

Person of the Month

Quiet excellence that holds the pipelines up

MK

Murali K.

DATA ENGINEERING · IDS · LILLY

"The right S3 path saves you a week. The wrong one costs you three. Get the contract right first, then build."

Automation

Jira → Excel in Three Phases

How we worked around the DLP block

◆ Automation · Internal Tooling

jira_to_excel.py — Direct API Access from SageMaker

Phase 1: discovered DLP blocks the obvious connector. Phase 2: provisioned a Jira Cloud API token for lilly-jira.atlassian.net (April 2026). Phase 3: hit the `/rest/api/3/search/jql` POST endpoint with `nextPageToken` pagination directly from a notebook. ES sprint splits and full project exports now run in under a minute, no manual UI work.

jira_es_splitter.py jira_to_excel.py IBUNBEES Direct API · Token Auth

Looking Ahead