Hello! My name is Joseph Moulton (or Joe, for short) and I’m a software engineer living in Boston Massachusetts. I grew up in Connecticut and have been living in the New England area all my life, don’t plan on leaving.
I will use this blog to occasionally post about projects I’m working on or topics I’m interested in. Not limited to software engineering or tech.
Also, quick note — I am not a web developer as made evident by the shitty design of this site 🙂.
</About my work>
At the time of writing (3/8/2026), I have been working in the tech space for about 4 years, all this time spent at Amazon. I am predominantly a backend engineer with experience working on both low level AI products as well as large-scale Java services.
The current product that I work on is Amazon’s Nova Forge, Forge is Amazon’s customer facing AI model customization service. Nova Forge offers a wide range of tools to train, evaluate, and host a custom model for inference. Currently, all customization tools are built to modify Amazon’s Nova model for a customers business use case. We offer different training techniques (like Supervised Fine Tuning, Reinforcement Fine Tuning, and Continuous Pre-Training) to tailor Nova’s performance to meet whatever use case a customer has. We also a wide variety of evaluation benchmarks which can be used to test the efficacy of the training. The customized models can then be deployed to AWS Bedrock for inferencing.
The specific subset of tools I work on developing within this suite are predominantly the evaluation and inferencing tools. I frequently implement new benchmarks and evaluation workflows to help customers understand and verify that their training worked.
A recent feature I helped launch within Nova Forge is the rubric based LLM-as-a-judge, which enables customers to use a custom trained judge model to numerically rate their model’s output. This provides quantitative values for which customers can judge their model’s performance based on the judge model’s generated criteria. See the blog post we published on this feature here: Evaluate generative AI models with an Amazon Nova rubric-based LLM judge on Amazon SageMaker AI (Part 2)
Before working on Nova Forge, I worked on Amazon Alexa’s end user data store — specifically on the data ingestion path. The service I worked on was a massive datalake which contained exobytes unstructured data corresponding to the output of every customer’s interaction with Alexa worldwide and subsequent data processing done by upstream producers. This data was used to feed downstream machine learning systems for improving Alexa’s internal model. During my time on this service I helped to shift our data ingestion architecture from the classic non-LLM based write patterns to the new transformer based LLM style data. This was done to power the new Alexa+ platform which greatly expands Alexa’s capabilities and allows for customer discourse with Alexa in a more traditional LLM style.
The Alexa datalake service was absolutely massive in scale, as we typically saw millions of requests hit our endpoints per second and our SLAs required 99.999% uptime. We stored incredible amounts of data as a result, with some of our S3 buckets being the largest in their respective regions.
Working on such a massive service was a great first introduction to the world of software engineering and distributed systems. I had to get very familiar with load testing, performance metrics, distributed systems, database architecture, load balancing, and just about every aspect of a large scale system you can think of.
</outside work>
Outside of work, I like to keep up to date with the latest developments in tech and gaming. I spend a lot of time and money building my computer setup which I use for both gaming and general development projects.
My computer specs are:
Internal specs:
CPU: AMD Ryzen 9800x3D
GPU: Nvidia RTX 5090 w/ undervolt and OC
Motherboard: Asus Tuf x670-Plus wifi
RAM: 32GB 6000Mhz CL30
Cooler: LianLi GA II TRINITY PERFORMANCE
Case: Corsair 5000D
Peripherals:
Mouse: Asus ROG Keris II Ace Wireless
Keyboard: Epomaker Galaxy100
Main monitor: MSI MPG 321URX - 32' 4k Oled 240HZ
Second monitor: LG UltraGear 32' 1440p 260HZHere are some pictures (ignore the dirty floor):
I predominantly play online games like World of Warcraft and Counter Strike. I run a gaming YouTube channel where I stream and hope to start making videos again (I used to back in like 2015-2018). Take a look at my channel:
I also have a golf addiction. I primarily play at Twin Hills Country Club and Blackledge Country Club in Connecticut.
Here’s some pics of my golf clubs that cost way too much money:
I also enjoy both listening and making music (moreso listening, I’m pretty shit at making).
I’ve been running the Bangers of the Month playlist series where I add music I’m listening to at that time in monthly playlists. I’ve been running it since Feb. 2018. See them here: https://open.spotify.com/user/killersharkdude/playlists.
Here’s my soundcloud where I post random stuff I make https://soundcloud.com/jguy25/tracks