JERUSAN
JEGA.
Software Engineer · Minneapolis, MN
Building systems for
messy, real-world data.

Right now I'm focused on using LLMs to turn unstructured compliance and operational inputs into structured, auditable workflows.

Email Me GitHub ↗ LinkedIn ↗
50K+
Daily actions processed
80%
Infrastructure cost reduction
7.5×
Faster vector search
<1s
AI eval latency
TypeScript✦ Go✦ Python✦ C++✦ AWS✦ LLMs✦ HNSW✦ Docker✦ PostgreSQL✦ React✦ Next.js✦ Node.js✦ TypeScript✦ Go✦ Python✦ C++✦ AWS✦ LLMs✦ HNSW✦ Docker✦ PostgreSQL✦ React✦ Next.js✦ Node.js✦
01 — Selected Projects
01
React . TypeScript · Python · Qwen3
AUDIT
GRAPH

Replaces weeks of manual auditor work with a reproducible, AI-assisted engine — without sacrificing the traceability compliance demands. Reconciles identities across HR, IAM, and ticketing systems; every finding is traceable, every workpaper replayable.

100%
Reproducible outcomes
0
Black-box decisions
02
C++ · ARM NEON SIMD
AS
VEC

Built from scratch on Apple Silicon — 489 QPS, 98.95% recall, 7.5× faster than exact search. ARM NEON SIMD vectorization with SoA memory layout and a 65K mutex pool for deadlock-free parallel index builds.

489
QPS on SIFT1M
7.5×
Over exact search
98.95%
Recall@100
03
Python · TypeScript · Next.js . Gemini-1.5-live
DEX

A voice-based AI interviewer that evaluates code in sandboxed containers — sub-second latency, no human in the loop. Real-time LLM feedback on algorithm design and code quality across multi-stage technical assessments.

<1s
Eval latency
0
Humans in the loop
02 — Experience
Midpoint Labs
Aug 2022 – Mar 2023
San Francisco, CA
Founding Engineer
  • Slashed external service costs by 80% — replaced third-party API dependencies with a custom listener-based blockchain integration layer that improved data reliability and gave the system full control over transaction timing.
  • Gave every stakeholder a live window into the blockchain — end-to-end request tracking from inbound pickup to final block confirmation eliminated blind spots and turned "where is my transaction?" from a support issue into a dashboard.
  • Stopped outages before they happened — a proactive health monitoring service replaced reactive incident response, automatically detecting failures and alerting developers before users felt a thing.
Diffuse Funds
Jan 2022 – Jul 2022
Chicago, IL
Software Developer
  • Improved trade execution speed and success rates in a highly competitive market — built a high-performance crypto trade engine that landed transactions for both human traders and algorithmic bots into the earliest available block, reliably.
  • Turned a several-day compliance burden into a 10-minute background task — automated the monthly trade audit reconciliation process end-to-end, eliminating manual filing entirely while improving accuracy across thousands of transactions.
KLDiscovery
Mar 2019 – Jan 2022
Eden Prairie, MN
Software Engineer
  • Unlocked a new tier of enterprise clients — led 7 engineers to overhaul the platform's permission system with granular backend security controls, making the platform viable for large law firms that previously couldn't adopt it.
  • Removed onboarding friction entirely — automated repository creation eliminated the support ticket workflow for new cases, letting law firms go from signed contract to active case without ever contacting engineering.
  • Moved the company upstream in the sales cycle — integrated pre-discovery cloud storage with a secure client-facing UI, enabling KLDiscovery to capture customers before discovery even began and expand its addressable market.
03 — Technical Skills
Languages
TypeScript Python Go C++ C# SQL GraphQL React Node.js
Infrastructure
AWS GCP Docker PostgreSQL CloudFront Message Queues
AI / ML
LLMs RAG Vector Search HNSW
Systems
ARM NEON SIMD Event-Driven RBAC
04 — Contact

LET'S
BUILD
SOMETHING.

Open to full-time roles, founding engineer opportunities, and interesting hard problems.


Contact
GitHubgithub.com/jerusan
↗
LinkedInlinkedin.com/in/jerusan
↗