Andrew Shamis

Applied ML Research Engineer

London, UK · andy@sham.dev · andrew.d.shamis@gmail.com

Abstract

I work on applied machine learning for drug design at Isomorphic Labs, building evaluation infrastructure that spans a portfolio of therapeutic programs. Before that I spent three years at QuantCo turning exploratory healthcare research into production inference systems. Oxford MMathCompSci.

Keywords: Algorithms, Machine Learning, Data Science

1 Experience

  1. 1.1
    Isomorphic LabsResearch Engineer, Applied ML, London, UKFeb 2025 — present · Python, SQL, JAX
    • Designed and implemented a general pipeline for targeted model evaluations against the portfolio of drug design programs, feeding into the workstreams of several teams.
    • Acted as ML point of contact for two critical projects collaborating with Eli Lilly, responsible for critical revenue streams for the overall business.
    • Pushed research on applications of IsoDDE / AlphaFold to novel capabilities in modelling compound behaviour with regards to novel bioactivity mechanisms.
  2. 1.2
    QuantCoMachine Learning Engineer, London, UKSep 2022 — Feb 2025 · Python, SQL, Rust
    • Sole engineer involved in launching a new ML product for steering the health outcomes of patients during medical leave.
    • Worked as one of two engineers within a team of 10 researchers implementing new approaches to survival analysis and diagnostic sequence prediction within the medical domain.
    • Migrated a 30,000 LoC exploratory ETL, training and evaluation pipeline into a deployable inference product within 3 months.
    • Designed and implemented a PoC for a distributed analytics environment, including provisioning of worker nodes for horizontally scaled CI/CD and ETL jobs, on tenanted client infrastructure.
    • Had sole responsibility for the reliability of a medical fraud detection service worth $40 million annually to the client.
    • Produced highly optimised rewrites of business and model critical code in Rust.
    • First software engineer in QuantCo's London office, through a period of large growth in the organisation's technical demands.
  3. 1.3
    Jane Street CapitalSoftware Engineering Intern, London, UKJun 2021 — Sep 2021 · OCaml
    • Developed a frontend service for reducing portfolio risk, by more effectively limiting trades performed by proprietary trading strategies on a high frequency client facing private exchange.
    • Optimised the runtime of a trade settlement system by redesigning it to operate as a pub-sub architecture over a distributed network with Apache Kafka.
    • Learnt a proprietary dialect of OCaml to a professional standard within a week.
  4. 1.4
    PalletFull Stack Engineer (Part Time), Oxford, UKSep 2019 — Aug 2020 · Python, TypeScript
    • First engineer besides the CTO at a web startup since valued at $100 million, building the early data infrastructure and web client.
    • Worked directly with the founder to create the first draft of a web frontend for a job searching platform.
    • Introduced ETL workflows to feed live web data to a novel job search algorithm.
    • Helped onboard new engineers before returning to my degree.
    • Contributions to the product helped raise $4 million in the first funding round.

2 Education

2.1
Oxford UniversityMMathCompSci — Master of Mathematics and Computer ScienceOct 2018 — Jun 2022
  • Machine Learning74
  • Advanced Machine Learning73
  • Artificial Intelligence74
  • Algorithms and Data Structures70
  • Computational Learning Theory74
  • Computational Game Theory73
  • Bayesian Statistical Probabilistic Programming72
  • Quantum Information78
  • Computer Security72
2.2
DissertationDesigning Interpretable Models using Neurosymbolic Methods81, Distinction
  • Developed a novel DL micro-architecture analogous to classical MLPs, designed with structural interpretability.
  • Linear layers are replaced with Schweizer-Sklar logical operations, which bound the domain to [0, 1] and have provably superior differentiability characteristics to existing solutions.
  • The proposed class of models can be effectively used for “soft-learning” of complicated binary decision problems. As interpretability is structural, the learnt policy can be expressed as a series of logical implications.
  • Demonstrated effectiveness at explaining the classification of MNIST images by reducing them to compositions of fundamental image components.

3 Projects

  1. 3.1
    Regex ParserRustNov 2020
    • Created a macro to parse regular expressions into NDFAs at compile time, to optimise the parsing stage of more sophisticated compilers.
    • Contains all patterns present in JavaScript's internal RegExp implementation.
    • Can recognise and isolate sub-expressions, for use in tokenisation.
  2. 3.2
    ParliaMateCo-Founder — Python, Android StudioNov 2018
    • Won the Oxford Hackathon 2019, with an app designed to help with interacting with UK Parliament proceedings.
    • Produced a service to pull live audio from parliament proceedings, and render annotated subtitles within the app, using the Google Speech-to-Text API.
    • Won the 'Deloitte API' and 'Google Cloud Compute Platform' challenges.