Find the right use
Where ML will actually help your users or your margins, and where a simple rule will do.
AI & machine learning · Nitish Arora
AI and machine learning, built into your product
A model is only useful once it's inside a product people use. I've built the whole path: collecting the data from apps and sensors, cleaning and processing it, training the model, and running it in the cloud or on the device. SleepDoc+ used radar signals and TensorFlow to stage sleep, and won the grand prize at the first Accenture Innovation Challenge.
- 10+
- years
- 8
- products shipped
- 200+
- people led
- 21
- press features
What you get
What I'll do for you
Data pipelines
Collect, label and store the data the model needs, from apps, APIs or hardware sensors.
Signal processing
FFT, filtering and feature extraction for sensor, audio and time-series data.
Models
Train and evaluate models with TensorFlow and Python, measured against a clear target.
Deployment
Serve models from the cloud or run them on the device, with monitoring once they're live.
Product analytics
Measure what users do with the feature, so the model keeps earning its place.
Proof
Products I've built that show it
My roleTech Lead (Phyzio × mBreath)
Contactless radar that screens for sleep apnea.
- Grand prize, Accenture Innovation Challenge
- Contactless
- 1,000+ data points
Featured in Accenture Newsroom · The Times of India · NDTV Education · The Tribune · CIOL · DQ Channels · BioSpectrum India · EducationWorld · Enterprise IT World
UWB radar sensing, STM32 and TI CC3200 firmware, FFT signal processing, sleep staging with TensorFlow and the companion app, across 1,000+ data points.
- UWB radar
- STM32
- TI CC3200
- FFT
- TensorFlow
- AWS
My roleCo-founder & Tech Lead
A wearable that relieves pain and fixes posture.
- Hardware to app store
- Indiegogo campaign
- Site in 7 languages
Featured in NS Business · MedX Tech
The device electronics and firmware, the iOS and Android apps with personalised therapy programmes, and the cloud behind them.
- Nordic BLE
- TI CC2541
- Java
- Objective-C
- Realm
- Django
- AWS
In the press
Featured in 21 outlets
- ForbesMissPoppins
- EntrepreneurMissPoppins on Elevator Pitch
- The Times of IndiaSleepDoc+
- ThePrintDhanLAP launch
- PR NewswireDhanLAP launch
- NDTV EducationSleepDoc+
- Accenture NewsroomSleepDoc+ grand prize
- Kitces AdvisorTech MapCherishBuckets
- ParentsMissPoppins
- MotherlyMissPoppins
- The TribuneSleepDoc+
- CafemutualDhanLAP
FAQ
Common questions
Do I need machine learning?
Often not at first. I'll tell you honestly where a model adds value and where simpler logic is faster and cheaper.
Can you add AI features to an existing product?
Yes. I start from your data and your users, then add the pipeline, the model and the product feature around it.
Have you shipped ML in production?
Yes. SleepDoc+ staged sleep from contactless radar signals using FFT signal processing and TensorFlow, and won the grand prize at the first Accenture Innovation Challenge.
Other services: