Developed an ML based product from idea stage to production deployment across 2 cloud platforms. This included internal pitches for funding at KPMG with the seed investment from the firm £250k with further investment in progress.
Problem Statement: It is time consuming and difficult to properly classify a company’s trading activity. Within banking, knowing what your customers do and where they operate is key to managing financial crime. A company’s trading activity exposes their banks to a certain level of risk and drives the majority of the due dilligence requirements. Additionally, this is not a point in time problem as a business banking customer could change their main trading activity at any point in time without informing their bank.
The product is a Machine Learning based alternative to understanding customer trading activities using online digital traces and extrapolating the risk they provide to a Bank. Uses Natural Language Processing to emulate analyst lead investigations.
The product is made available as a subscription based API that banks can integrate into different product where understanding a customers industry and risks are essential (client onboarding, due dilligence, periodic customer review etc.)
The team deployed 2 ML models and 16 Heuristic Models to production, passing the models and ecosystem through the internal Model Risk Process.
Developed an NLP Model (BERT-Based) fine-tuned on Customer Due Dilligence (CDD) domain language to extract key information from unstructured text via both NER and Question Answering tasks.
Built Machine learning models for a large UK bank, focusing on streamlining the complaints process. Developed an NLP classifier to judge which complaints could be accelerated through the process. Built an anomaly detection algorithm to warn of sudden spikes in volumes and link back the spike to a root cause.
Text Based ML
Predicting Brexit using Twitter as a high frequency polling mechanism. Built a fine-tuned DistilBert model to predict whether a spefic tweet was for brexit (anti-EU), against brexit (pro-EU) or neutral. 🔗 Presentation
Image Based ML
Using YOLO for CCTV monitoring of accessibility assets (i.e. lifts and escalators) in train stations. The pipeline counts the number of passengers using each asset over the course of the day to optimise the maintenance cycles used by Network Rail. 🔗 Write-up in Progress
Predicting Brexit using Twitter as a high frequency polling mechanism. Built 3 Image Classification models using an EfficientNet based model to identify the person’s demographics (age, race, ethnicity). The demographics tied in with the political opinion and stated profile location was used in conjunction with the national census to create representative opinion distributions across the country.
Generative ML
Generative Radio is a submission for Kontinuum 2022, which combines the changing seasons throughout the year with a well known composition by Vivaldi (Four Seasons) to generate a unique and ever evolving composition which will play for 365 days in a row if selected by the panel. The musical generation is supplemented with artificially generated narration, which is programmatically triggered to play every hour, similar to a radio presenter. 🔗 Write-up in Progress
Persigan is a fine-tune Image based GAN model to generate new Persian Carpet designs 🔗 Write-up in Progress
Reinforcement Learning
Cryptocurrency trading using the TensorTrade RL framework to trade cryptocurrencies. 🔗 Write-up in Progress
Designed then lead a team of 8 to build a customer data monitoring system for a Tier 1 Bank’s high risk customers. Built in Python with 3rd party API integrations, reporting built in Pug.js
Designed then lead a team of 7 to build a configuration based migration tool to transform data from legacy systems (relational) into a single customer view as a JSON. Developed a full customer decision tree which identified any missing data that is required as per policy tailored to every customer’s circumstance.
InfraHack 2019 (1st Place)- Lift & Escalator Optimisation and Real-time Data Feeds using CCTV
HackTrain VI (Finalist, Winner SilverRail Challenge) –Production grade chatbot in DialogueFlow