Buddhika Seram

Head of AI Engineering

Blend

Jan 2025 - Current (1 year 9 months)

Leading AI engineering across Blend360's EMEA practice. My remit is
architecture authority across active engagements, technical direction
of senior leads and workstream owners, and being the escalation point
when the team hits something they can't resolve. At this level I'm not
in the codebase — I'm the person who shapes the decisions that are hard
to undo and signs off on the ones that matter.

Two things I'm most proud of from this period:

Trellis IQ - Blend360's commercial agentic AI platform for CPG data
harmonisation. Processes 550 records per minute at 98% precision across
7 languages. Recently became a Databricks GenAI Partner Accelerator.
I'm named on the product p

Lead Data Engineer

Blend

Apr 2023 - Jan 2025 (1 year 10 months)

This is the period where I built Blend360 EMEA's first
software-engineering-grade AI products - not dashboards or notebooks,
but Dockerised FastAPI applications with CI/CD pipelines and real users
in production.

The headline work was an AI suite for a global med-comms client - three
interconnected applications spanning automated literature search and
summarisation, meeting intelligence for field teams, and AI-assisted
content generation. The engagement won Innovation of the Year at the
British Data Awards 2025 alongside our client.

On the first application I was fully hands-on. It ran end-to-end in 15
minutes on first completion against a 5-minute SLA - spent a week on

Senior Data Engineer

Blend

Oct 2022 - Apr 2023 (7 months)

Joined Blend360 as the AI and MLOps specialist within a data engineering
practice. Early projects included Comcast - where I came into a failing
offshore delivery, stabilised it, and led the migration from Scala to
PySpark. Also built Databricks bronze/silver/gold pipeline templates that
became reusable across the EMEA practice.

This was the period where I understood what was missing from the team:
software engineering rigour. No IaC, no CI/CD, no repeatable deployment
patterns. That became the thread I pulled on for the next two years.

Research DataScientist II

ebot GmbH

Aug 2019 - Oct 2022 (3 years 3 months)

Three and a half years building production ML systems at an early-stage
conversational AI company - the kind of work where there was no existing
infrastructure and you had to design and ship everything yourself.

The infrastructure work: designed and built a Kubeflow-based continuous
training pipeline (CTP) that automated the full cycle from data
acquisition to model deployment. Built Catwalk - an inference
orchestration microservice that managed model serving across environments.
Migrated model serving to SageMaker on EKS, which cut pod CPU usage by
70% and reduced training times by 30%.

The serving work: ran systematic load testing on Triton inference server
and achieved

Intern Data Scientist

ebot GmbH

Mar 2019 - Aug 2019 (6 months)

i

Mechatronic Engineer

iMatrix Automation Pty Ltd

Jul 2013 - Aug 2018 (5 years 2 months)

Five years as a sole-trader subcontractor designing, programming, and
commissioning industrial control systems across Australia and
internationally. Full project ownership from design through to
on-site commissioning - no senior engineer to escalate to.

Projects spanned powder blending systems, drilling rig automation,
packaging and marking lines, checkweigher systems, and PLC-to-PLC
communication networks. Clients included Easternwell, Goodman Fielder,
Weir Minerals, and Frostyboy.

One that stands out: upgraded a checkweigher data acquisition system
from VBA to a multi-threaded Python application - my first real
production software deployment, before I knew that's what i

Thesis Student

Capral Aluminium

Jan 2013 - Jul 2013 (7 months)

Honours thesis project: designed and built an autonomous pick-and-place
system to replace a manual die cast library process. The system used
image processing and character recognition to identify and validate
extrusion die casts before placement.

First time I used computer vision in a real environment. The problem of
making reliable automated decisions under variable conditions - lighting,
wear, positioning - turned out to be a good introduction to what
production AI actually involves.

About

Detail

Timeline

Résumé

Buddhika Seram

About

Detail

Behavioral traits