Delivering AI Applications with Open Source Platforms at Scale

    Pantelis Monogioudis, Gurudutt Hosanghadi, Lorant Farkas, Chris White
    Augmented Intelligence & Devices Research Lab
    Bell-Labs, Murray Hill, NJ, USA

About me

20 years of experience in leading Research & Development projects. I started in wireless communications (PHY/MAC) in the 90s and in 2000’s started diversifying into Machine Learning initially for wireless networks and later for all sorts of applications. Currently leading ~15 researchers in Murray Hill NJ and Budapest in Applied Machine Learning working on real-time multisensory machine learning applications. 1 (failed) startup, 15 journal publications and 50 patents issued/pending. As of spring 2019 I am also with the Computer Science department of NJIT. I live in Randolph, NJ with my wife, my two boys Alex & Nicholas and my dog Bleu.

We are hiring ML researchers / engineers in Paris-Saclay region - please get in touch with me via linkedin / twitter if you are interested.

A nautical analogy on where we are today with AI

Cumberland Basin, April 1844

DARPA's AI development in waves

Symbolic GOFAI

> 1980


> 2010

AGI (kind of)


Meaning and resoning in NLP, Inference and Representation of Causality, Uncertainty Representations, Long-term Goal Planning, World Models

AI in Health

Capturing the AI opportunity - case study

The team and the idea behind it

The Business Model

The Architecture

The Implementation (Pipelines)

The Market Validation & Exit

640,000 people die every year

Yearly cost of heart disease treatment: $300 Billion

The Idea

Generative augmented physical (CFD) modeling from CT Scans

Up to now the determination of blockage involved threading a catheter from the groin up to the heart and measuring blood flow to determined whether a stent is needed to open up the coronary artery.

The Team

Medical Doctors

Mechanical Engineers

Computer Scientists

NOTE: No FDA Approval Required (e.g. 510K clearance)

The Business Model

Medicare reimburses Heartflow $1,450 per test.

For every reimbursement Heartflow saves $4,000 in costs

The test is non-invasive and therefore almost ambigutous.

The company has partnered with SIEMENS' Marketplace and other vendors/providers.

What we need from AI Architecture

  • Composability

    pointing to microservices and atomic operations.
  • Scalability & Resilience

    pointing to cloud-native (Kubernetes and FaaS) architectures.
  • Portability

    across workstation, training rigs, dedicated edge devices and public clouds.

AI Architecture Building Blocks - Deep Learning Foundation

AI Stack

The significance of the edge

The 4 pipelines (Plumbing)

Data Pipeline

Model Training Pipeline

Model Evaluation & Validation Pipeline

Serving Pipeline - the need for AI marketplaces

Market Reaction

  • The single most important success factor of explainable predictions
  • 2019 market valuation - $1.5 billion
  • NHS adopted the approach in all its hospitals

What other opportunities are there ?

Video AI opportunities in surveillance

Deep Learning in 2017

After spending 3200W, 2-socket CPU, 8 x GTX1080Ti, 16 x 720p cameras, algorithms are unable to pass the mirror test.

2 year old

She figures it out in less than 1 min.

Video AI opportunities in sports

Lots of opportunities on improving team performance or AI-based personal trainers.

$200 for 1 camera

$3,000 to store in AWS S3 its video feed for 1 year

Race to Annotate

Video AI opportunities at the edge

  • All cloud providers are working in capturing hybrid cloud native applications hosted on Kubernetes

  • Federation, tailored to the edge hardware, security & other optimizations for edge clouds and embedded systems


Autonomous connected electric & shared (ACES) vehicles
  • How will EVs recharge overnight in a city full of blocks of flats?
  • Simoudis' book covers the business models and technology drivers

Congestion pricing in city centers
  • Yield Management Algorithms (Dynamic Pricing)
  • License plate recognition
  • Car type recognition
  • Traffic Congestion recognition

Closing ...

  • There is no such thing as "AI startup" - you need to have a solid business model that is by itself sustained without AI
  • Rethink mission critical industries - partner with local providers to make their business models work better
  • Health, Transportation, Energy, Security are being disrupted not so by AI but by the drammatic cost reduction of the software & hardware platforms that allow entries from many, unseen before, players

Thank you!