$

Hyemin Kim

AI Solutions Engineer

I design AI systems in the language of the customerand understand customers in the language of technology.

SCROLL
// 01

About

I sit at the intersection of AI engineering and customer-facing delivery.

At Squarelight, I own the full lifecycle of enterprise AI PoCs — from the first client meeting to architecture design, hands-on development, and final deliverable handoff. I translate business pain points into working AI systems, and complex technical concepts into language that stakeholders trust.

My background spans NYC business development, Google ML training, and bilingual communication (Korean/English, OPIc AL). This combination lets me move fast between boardrooms and codebases.

I'm looking for a role where I can keep doing what I do best: building real AI solutions with customers, not just talking about them.

AI / ML

LangGraphLangChainRAGPrompt EngineeringClaude APIOpenAI API

Backend

PythonFastAPIPostgreSQLQdrantWeaviaten8n

Frontend

Next.jsTypeScriptReactTailwind CSS

Pre-Sales

PoC Design & DeliveryTechnical ProposalsSoW AuthoringClient CommunicationRequirements Engineering
// 02

Experience

Work

Squarelight

AI Solutions Engineer

Jan 2025 — Present · Seoul

  • Lead client-developer communication for enterprise AI PoCs end-to-end
  • Architect and deliver AI solutions for clients including Shinsegae Department Store and Jinsol Customs Corp.
  • Author all client-facing deliverables: technical proposals, architecture docs, PoC reports, and SoW documents

Very Good Ventures

Business Development Intern

Oct 2019 — Jan 2020 · New York, NY

  • Managed 10,000+ record datasets for business intelligence and sales strategy
  • Co-organized Flutter developer event with Google for 200+ attendees

Absolute New York

Account Coordinator

Apr — Oct 2019 · New York, NY

  • Managed 400+ B2B wholesale accounts across North America
  • Streamlined invoicing and shipment tracking to reduce order discrepancies

Education & Training

Woori FIS Academy

AI Engineering

Jul — Dec 2024

Google ML Bootcamp

Machine Learning & Deep Learning

Sep — Nov 2023

Sangmyung University

B.A. English Education / B.A. Global Business

2015 — 2021

Languages & Certifications

OPIcAL
TOEIC985
Deep Learning SpecializationCoursera
// 03

Projects

Enterprise AI PoCs and systems I've designed, built, and delivered — from concept to production contract.

AI Ad Content Generator

2025

Automated Ad Creation System — Enterprise PoC

AI system that auto-generates marketing copy, background images, and product cutouts for department store ads.

~20s
Per Page Generation
8 Types
Page Templates
Hours → Sec
Time Saved

My Role

Led all client-developer communication as sole technical liaison. Drove requirements gathering, service feature design, and authored all technical deliverables.

What I Built

Enterprise PoC for Shinsegae Department Store's marketing team. The system generates complete ad pages — marketing copy, AI-generated background images, and product cutouts — across 8 different page templates in approximately 20 seconds, replacing a multi-hour manual workflow.

Tech Stack

AI Image GenerationPrompt EngineeringEnterprise PoCTechnical Documentation

Architecture

Page Template
8 Types
Product Data
Images + Info
AI Engine
Content Generation
Marketing Copy
AI Generated
Background
AI Image Gen
Product Cutout
Auto Extraction
Composition
Layout Engine
Ad Page Output
~20s per page

Customs AI System

2025

LLM-Powered Customs & Compliance — Jinsol Customs Corp.

AI customs consultation system with RAG-based legal Q&A, HS code classification, and import anomaly detection.

8
Core AI Features
Solo
PoC Development
Contracted
Multi-phase Production
Legal Q&A
Legal Q&A1 / 6

My Role

Sole planner and developer. Owned everything from PoC architecture to client-facing documentation and contract negotiation support.

What I Built

Solo-built PoC that validated 8 core AI capabilities: RAG-based tariff law Q&A, automated HS code classification, import declaration anomaly detection, customs drawback analysis, and more. Authored the SoW and all project documentation that directly contributed to securing a multi-phase production contract.

Tech Stack

RAGLLMHS Code ClassificationAnomaly DetectionSoW Authoring

Architecture

LLM Core Engine
Central Processing
Tariff Law Q&A
RAG
HS Code
Classification
Import Anomaly
Detection
Drawback
Analysis
Regulation
Search
Document
Analysis
Compliance
Check
Consultation
AI Chat
Legal DB
Tariff Laws
HS Code DB
Classification
Case History
Past Decisions

OARIA

Dec 2025 — Jan 2026

Oncology AI Research Intelligence Assistant

Agentic RAG system that lets researchers query, synthesize, and get evidence-based answers from oncology papers.

3-Gate
Safety Verification
Hybrid
Dense + Sparse Retrieval
Bilingual
KR / EN Zero-shot

Add slides / demo video

/public/projects/oaria/

My Role

Sole architect & developer — designed and built the entire system end-to-end.

What I Built

Built a full agentic RAG pipeline from scratch. The system uses BGE-M3 hybrid retrieval (dense + sparse vectors) with Qdrant, LangGraph-based agent that decomposes complex multi-hop questions into subtasks, and a 3-stage safety gate: domain classification → retrieval confidence → RAGAS faithfulness scoring. Supports real-time SSE streaming of the agent's reasoning process.

Tech Stack

LangGraphFastAPIQdrantBGE-M3Claude APINext.jsTypeScriptPostgreSQL

Architecture

User Query
KR / EN
Gate 1
Domain Classification
LangGraph Agent
Task Decomposition
BGE-M3
Dense + Sparse
Qdrant
Hybrid Search
Gate 2
Retrieval Confidence
Claude API
Answer Generation
Gate 3
RAGAS Faithfulness
SSE Stream
Real-time Response
PostgreSQL
Metadata
Qdrant
Vectors
FastAPI
Backend
Next.js
Frontend

AI Content Pipeline

Jul — Sep 2025

Automated News-to-Instagram Publishing

End-to-end automation: news collection → AI content generation → image rendering → social media publishing.

5hr → 5min
Time Reduction
Daily
Autonomous Operation
E2E
Fully Automated

Add slides / demo video

/public/projects/pipeline/

My Role

Built the entire end-to-end automation pipeline.

What I Built

Built an n8n-based workflow that collects news via RSS, clusters and selects topics through multi-stage Claude API prompt chaining, generates card-news content, renders images with Puppeteer, and auto-publishes to Instagram via Graph API. Runs autonomously every day.

Tech Stack

n8nClaude APIPuppeteerInstagram Graph APIGoogle Sheets APIJavaScript

Architecture

Google News
RSS Feed
n8n Workflow
Orchestration
Claude API
Multi-stage Prompts
Topic Selection
AI Clustering
Content Gen
Card News Copy
Puppeteer
Image Rendering
Instagram
Auto Publish
Runs autonomously every day
// 04

Let's Connect

Looking for an AI Solutions Engineer who can bridge the gap between technical possibilities and business outcomes?

© 2026 Hyemin KimSeoul, South Korea