Alex Hall Email & Phone Number

London, England, United Kingdom

New York, United States

Part of a small team building Grist (https://www.getgrist.com/), a web application that's a hybrid between a spreadsheet and a database.- Reduced page load from 4 to 2 seconds by optimizing Python packaging.- Migrated a large complex codebase from Python 2 to 3 while preserving backward compatibility, even with arbitrary user-written Python code.- Implemented a new webhooks system, allowing Zapier integrations to respond to events in seconds instead of minutes.- Replaced a complicated system of communication between components with a simpler implementation that was compatible with more browsers and had fewer restrictions.- Pioneered a new AI assistant to help users write spreadsheet formulas, now one of the most exciting new features.- Unified separate methods of data conversion into a single coherent system, resolving a major category of user complaints.- Allowed users to make HTTP requests in formulas, despite formulas being evaluated in a sandbox with no network access and needing to return results immediately. This enabled vast new possibilities to connect their documents with the outside world.

San Francisco, California, United States

Building tools to make it easier to combine many calls to LLMs (e.g. ChatGPT) to perform complex reasoning:- Replaced an off-the-shelf algorithm for clustering vectors with a novel custom algorithm which increased accuracy from ~50% to ~95%.- Migrated a complex application to a new tech stack and packaged it as an easily installable library.- Rearchitected the data layout to enable visualizing large files, which previously froze the application.- Reproduced and resolved a months-old issue that no one else had been able to solve, almost halving the overall rate in production.- Allowed visually distinguishing important components of templated strings without having to significantly alter code, which was only possible using my own library.

Cape Town Area, South Africa

Full stack engineer with a huge array of responsibilities, from maintenance of large old legacy systems to completing new projects from scratch, all in a small team. I wrote countless scripts, automated jobs, HTTP APIs, clients for other APIs, dashboards, and complex user interfaces. Some major achievements:- Reimplemented a computational engine to use AWS Athena instead of Apache Flink, which was simpler, cheaper, and took minutes instead of hours.- Developed a specialized mathematical system to look up values from a large table which was faster and more efficient than a SQL database.- Created a pipeline to reliably extract complex structured data from scans of bank statements, with a system to detect and correct errors made by the OCR.- Automated deployment of Docker containers in AWS ECS, saving time previously spent on a manual, tedious, and error-prone process.

Cape Town Area, South Africa

My main project here involved collecting data from Amazon SQS, transforming it, and uploading it to Google BigQuery in an efficient, robust, and idempotent manner. Other than that there were various bugfixes, code refactorings and enhancements on many parts of the system. The vast majority of work was in Python.Yola takes code quality very seriously and every change and decision was carefully considered. This taught me to how work better in a team and resolve disagreements in a productive manner, as well as how to make my code easier to read and maintain.

I have interned here twice. The first time was around December 2013, when I developed a web server with a RESTful API in Java which performed inferences using Bayesian Networks (machine learning). This involved:• Doing research to find and compare libraries, including evaluating their relative speeds.• Creating a server which handled different HTTP requests using a custom API with messageswritten in JSON, called the appropriate library methods to perform analysis, and interfacedwith AWS S3 to store models.• Learning statistical theory and applying it to write algorithms to solve some problems thelibrary did not solve.In 2015 I did a large variety of tasks of different sizes, such as:• Efficiently processing CSV files while streaming them from and to S3 in parallel. For this I created an open source library: see the attached link.• Fixing bugs in the app, which required working with Django, Celery, MySQL and Dojo.• Deploying a handler on AWS Lambda written in Scala which parsed and compared raw HTTPmessages stored on S3.• Experimenting with Apache Hive and Spark to improve performance for manipulating largetables of data.• Creating and using various Docker containers for tools including Riemann, Sentry, CeleryFlower, and crontab.

Worked on an ERP system, which included:• Various miscellaneous ops tasks, such as: - Deploying Elasticsearch, Logstash, and Kibana on an AWS instance and connecting it to the application and its existing logging code via Redis. - Packaging, configuring and deploying Riemann on another instance to gather metrics. - Patching an AWS CloudFormation template for Datomic to allow metrics to be sent to Riemann.• Integrating a system to mathematically forecast inventory demand into the app. The backend, algorithms, and evaluation of data was written in Clojure, while the frontend was written in ClojureScript using Bootstrap and React.

Wrote a number of complex tools to assist developers in Python:• A client which communicated with a Twisted manhole server over telnet through SSH tunnelsto execute arbitrary code in an RPC style to aid debugging on servers and the construction ofdiagnostic scripts.• A tool to thoroughly analyse the state of a cluster’s DNS system, find problems, and suggestsolutions. This included inspecting the database for inconsistencies, performing DNS queries and zone transfers, checking network connections and firewalls, searching logs, and parsing configuration files.• A tool which ingested large amount of arbitrary textual logs and organised them into groups sharing a common structure with a few varying parameters using a complex novel algorithm, allowing developers to see a summarised version of what can otherwise be an overwhelming amount of data.

Sentiment analysis using various machine learning classification methods in Java and Groovy. This involved:• Understanding the current code used to solve this problem, and fixing various issues it had.• Processing natural language text using regular expressions to extract essential informationdespite variations in spelling, punctuation, etc.• Understanding the natural language processing library MALLET, the various classificationalgorithms it offered which could be applied to the problem and how to use and configurethem, and how to extend it with my own algorithms.• Trying out other libraries to solve the same problem.• Setting up an elaborate framework to compare the accuracy and efficiency of differentcombinations of classification algorithms, solutions to subproblems, and configurationparameters, and using it to find the optimal way to solve the given problem.

Developed a plugin in Ruby (with a JavaScript/HTML interface) for the 3D design programSketchup to fit grids to surfaces and analyse the amount of sunlight hitting nodes over a year. See the attached link. This involved:• Learning Ruby and Javascript for the first time.• Using the Sketchup API (Ruby), which was often poorly documented or even buggy andrequiring workarounds.• Producing a large amount of data via geometrical calculations (Ruby).• Designing and implementing a dynamic user interface (Javascript) for the analysis, with thehelp of a few libraries including the SVG library Raphaël.

Development of a social networking Android app, involving:• Learning a lot of Android programming, especially to create user interfaces for different sections of the app.• Understanding a fairly large amount of existing, sometimes poorly written code to which features needed to be added.• Sending to and receiving messages from a server via a custom protocol on top of TCP, thus having to keep track of a large number of different kinds of asynchronous callbacks.• Collaborating with my fellow intern, including merging code and discussing design.