James Barnden

- Built and maintained web crawlers for harvesting 10.1 million raw job postings for processing
- Constructed a modular pipeline for the ingestion of raw postings, working closely with the machine learning team to adapt their trained models for scale.
- Implemented a duplicate removal module to handle processing at scale, reducing the corpus to 6.8 million unique postings.
- Collaborated with machine learning team on a scalable field normalization module.
- Identified pipeline… Show more

- Built and maintained web crawlers for harvesting 10.1 million raw job postings for processing
- Constructed a modular pipeline for the ingestion of raw postings, working closely with the machine learning team to adapt their trained models for scale.
- Implemented a duplicate removal module to handle processing at scale, reducing the corpus to 6.8 million unique postings.
- Collaborated with machine learning team on a scalable field normalization module.
- Identified pipeline bottlenecks and improved ingestion speed by 80%, reducing the time to process the dataset from 2 weeks to 48 hours. Show less

• Solo full stack development of an online multiplayer game app with real time events synced across multiple devices.
• Low/no cost (free tier) fully serverless autoscaling infrastructure across AWS and GCP.
• Python FastAPI backend/API, running on AWS Lambda & API Gateway.
• React PWA deployed to S3 served via CloudFront CDN.
• Can also be packaged and deployed as a native Android app

A desktop application for the classification of texts into user defined categories, tested across multiple subject areas and users with strong results.

The application consists of a modular pipeline of interchangeable Data Science components made accessible to an end user for configuration and training via GUI. The pipeline handles data acquisition, feature selection, dataset creation, model training and classification, with an emphasis on Natural Language Processing. The pipeline can… Show more

A desktop application for the classification of texts into user defined categories, tested across multiple subject areas and users with strong results.

The application consists of a modular pipeline of interchangeable Data Science components made accessible to an end user for configuration and training via GUI. The pipeline handles data acquisition, feature selection, dataset creation, model training and classification, with an emphasis on Natural Language Processing. The pipeline can be extended to process other data formats such as images, video or audio. Show less

This project involved developing a series of plugins (data acquirers, data miners & search engines) for a configurable pipeline, to construct a series of searchable corpuses (e.g. to hold information drawn from audio, images, text and metadata) from a given video URL. The information held in each corpus is mapped to a timeline, such that a user can perform a search which returns one or more time segments of the video, relevant to the query.