Segment 01AI · Finance
SECDigest
An AI platform that turns 300-page SEC filings into financial dashboards and RAG-powered answers, running on a $0 infrastructure budget.
- Filing length
- 300 pp.
- Infra budget
- $0 / mo
Annual ReportFiscal year ending December 31, 2026
Erik Larson
Software engineer. Full-stack, backend and applied AI.
I build full-stack, backend and AI systems that stay fast, reliable and correct in production. Graduating from Georgia Tech in December 2026 and looking for full-stack, backend, platform or applied AI roles.
- Education
- Georgia Tech, B.S. Computer Science, minor in Economics
- Graduation
- December 2026
- Seeking
- Full-stack, backend, platform or applied AI roles
- Hometown
- Seattle, WA
Indicate by check mark whether the registrant is open to full-time roles starting in 2027.
Indicate by check mark whether the registrant has shipped code to production.
Indicate by check mark whether the registrant is a shell company.
Results, by the numbers
| Measure | Before | After | |
|---|---|---|---|
| Config API response timeCodeMettle, 2026. Redesigned per-asset rows into per-type manifests.20× faster | 5.00 s | 0.24 s | 20× faster |
| Database writes per pack applyCodeMettle, 2026. Same redesign; the before figure is approximate.−98% | 660 | 14 | −98% |
| Historical telemetry writes, relativeCodeMettle, 2026. Live-only routing for the real-time telemetry subsystem.8.1× fewer | 8.1× | 1.0× | 8.1× fewer |
| House-price model error (MAE)SoCal House Prices. XGBoost on 15,474 Southern California home sales.−36% | $228K | $145K | −36% |
- tests in the platform’s async suite
- 501
- endpoints moved off a manual copy workflow
- 34
- query-routing accuracy, FinMA research
- 86%
- monthly infrastructure for SECDigest and Vigil
- $0
A note from Erik
Dear reader,
I'm a computer science student at Georgia Tech with a minor in economics. I spent the last year on CodeMettle's systems team, first as an intern, then as a co-op.
My main project there was an internal developer platform I built from scratch. It replaced a manual copy workflow with an automated one: device template and panel packs are now assembled, versioned and deployed by the platform, every deployed asset traces back to its exact commit, and merging to main is the whole release. Along the way I took a core configuration API from 5s to 0.24s.
When my co-op ended, I kept building. SECDigest turns SEC filings into financial dashboards, and Vigil turns production alerts into Slack briefs. Both run publicly at zero cost.
Before that, I did NLP research at Georgia Tech's Financial Services Innovation Lab. Between that and my economics minor, I've picked up an interest in fintech.
Outside of code, I'm from Seattle and a big Seahawks fan. I also enjoy trying new foods, hiking, working out, and dabbling in photography.
Sincerely,
Erik Larson
Selected projects
Segment 02AI · Reliability
Vigil
An autonomous incident responder that turns Prometheus alerts into a likely-culprit commit and a Slack brief, using at most 3 LLM calls per incident.
- LLM calls per incident
- ≤ 3
- Infra budget
- $0 / mo
Segment 03Native app
Kapture
A native macOS photo booth that turns a burst of camera shots into a styled photo strip. Built with Swift and SwiftUI.
- Platform
- macOS
- Stack
- Swift, SwiftUI
Segment 04Machine learning
SoCal House Prices
An XGBoost model for 15,474 Southern California home sales that nearly doubled explained variance (R² 0.35 to 0.64) and cut mean error from $228K to $145K.
- Home sales
- 15,474
- R²
- 0.35 → 0.64
Things you should know
Hiring Erik involves a degree of risk. Prospective employers should consider the following.
- 1.
Erik measures things.
A 5-second API will be noticed, profiled and fixed. Expect numbers in pull request descriptions.
- 2.
Erik prefers boring releases.
Merging to main should be the whole deploy. Manual copy steps rarely survive the first quarter.
- 3.
Erik is a Seattle Seahawks fan.
Productivity on Mondays during football season may correlate with Sunday’s result.
- 4.
Erik may be away from the desk.
Off hours go to hiking, working out, trying new food and photography. Erik usually comes back with good photos.
Experience
- Undergraduate Researcher, Financial Services Innovation Lab, Georgia Tech, Aug 2023 – May 2025
- Software Engineer Intern, Product Development, CodeMettle, May 2025 – Aug 2025
- Software Engineer Co-op, Systems Engineering, CodeMettle, Jan 2026 – Aug 2026
CodeMettle
Software Engineer Co-op, Systems Engineering
- Architected an internal developer platform replacing a manual copy workflow across 34 endpoints, automating assembly, versioning, and deployment of device template and panel packs behind a 501-test async suite.
- Reduced a core configuration API's latency 20x, from 5s to 0.24s, and cut writes per pack apply ~660 to 14, by redesigning per-asset rows into per-type manifests.
- Automated the release lifecycle end to end, from semantic-version bumps and Kaniko builds to Helm publishing and auto-generated rollout MRs, making merge-to-main the whole deploy path.
- Shipped drift detection, diffing, and revert for versioned asset packs, so any deployed asset traces to its exact commit.
- Delivered a real-time telemetry subsystem raising device polling from 90s to seconds mid-incident, with live-only routing that cut historical writes 8.1x.
CodeMettle
Software Engineer Intern, Product Development
- Built a migration utility converting legacy Cytoscape network maps into React Flow topologies, then delivered 15+ features and fixes for the Terminus edge-node monitoring platform.
- Hardened LDAP authentication by normalizing attribute-key casing and converting silent role-resolution failures into explicit fail-closed denials, closing a privilege-misassignment path, backed by 240+ lines of new tests.
Georgia Tech
Undergraduate Researcher, Financial Services Innovation Lab
- Developed a multi-task model jointly learning event extraction and stock movement prediction from financial text, using fine-grained events as distant supervision (AllenNLP, NLTK, Stanza), trained on SageMaker.
- Engineered a financial Q&A application on a team of 5 routing queries across GPT-4 and FinMA variants via a fine-tuned BERT classifier reaching 86% routing accuracy, evaluated on 14 FLARE benchmark tasks.
Activity on GitHub
- contributions in the last year
- 443
- public repositories
- 9
- most-used language
- Python
Composition by language
| Python | 47.1% |
|---|---|
| TypeScript | 26.3% |
| Swift | 13.4% |
| JavaScript | 5.5% |
| HTML | 4.8% |
| CSS | 1.9% |
Share of code bytes across public, non-fork repositories. Top 6 shown.
Documents and links
Get in touch
Have a question or an opportunity? Send a message, or email erik.larson12@outlook.com.