Skip to content
EddieSun
home01writing02reading03

  • rss

  • github

  • linkedin

© 2026 Eddie

About.

Currently leading signal evaluation at a fintech.

Fields I find fascinating are decision theory, epistemology, and cosmology. My favourite authors are Deutsch, Isaacson, and Feynman (cliché).

I studied Systems Design EngineeringA co-op program combining software, mechanical and electrical engineering with design and human factors.x∗=argx∈Fmin​E[C(x,ξ)]x* equals argmin over x in F of the expected value of C of x and xi.Coursework ranged from modelling and optimization to statistics, pattern recognition and scheduling.See an example class profile at uWaterloo.

Some companies I worked on:

  1. Simulating atoms (Direct AtomicsWe trained a 13.8M parameter equivariant eSCN model on Alexandria's DFT data to estimate energies and forces from atomic structures.Screening materials faster wasn't enough. Semiconductor customers had billions invested in making their existing materials work. Read the post-mortem here.Comparison chart: traditional DFT scales O(n^3) and costs HPC hours-to-months for hundreds-to-thousands of atoms, while the neural network surrogate scales O(n) and runs in sub-second GPU time on a single GPU)
  2. Modern RelayAn agent works out how to find something in the company's data. Save that query so the next agent doesn't have to work it out again. Agents can propose changes to the data model too. New facts go onto branches for review, so mistakes can be discarded before they reach the shared graph. The company owns the data and its definition.Raised $3M USD. for agents and knowledge work.
  3. Helping websites show up in chatbot answers (λ eignWe tracked what chatbots said about brands, then worked on making their sites easier to read and cite.Wound it down as tracking tools became interchangeable and Cloudflare moved into markdown for agents.Domain rank distribution chart for thenetworkstate.com showing how often the domain is cited across GPT, Perplexity, and Claude responses, alongside example AI-cited sources)

Some companies I've worked at:

  1. Support ticket clustering at CloudflareBERT-based clustering for high-volume internal support tickets.Pivoted from topic clustering to finding similar historical tickets for technical support engineers.Result: 95% of tickets had useful similarities; about 50% of tickets were resolved 80% faster.BERT→UMAP→HDBSCAN→c-TF-IDF→MMREmbedClusterLabel (early 2022).
  2. Legal document A legal assistant that answers questions from PDFs. Extract the text, split it into chunks, embed them, then retrieve passages for the model to use.Explored GPT-4 and local models. at Protocol Labs (early 2023).
  3. Product/design/swe internships (BlackBerry QNX w/ AWS, Wind River Studio for cloud-native embedded systems, Mount Sinai cytopathology).

This is the latest snapshot of my personality archetype.

Some things I believe in:

  1. Mouth closed, don't get fed.
  2. Desperation never served anybody. Suffering is a choice.
  3. Life is too short to play zero-sum games (e.g. status).
  4. Kindness isn't niceness. Signaling virtue is a vice.
  5. Character is grace when the chips are down.
  6. Grit is a superpower, but so is quitting.
  7. Be a hyperrealistAppreciate every reality, even the harsh ones. Despise impractical idealism.Dreams matter, but the people who build great things aren't idle dreamers. They're grounded in reality, understand cause and effect, and use principles to get what they want.. Dreams + Reality + Determination = A Successful Life.