Seongwook Hwang

Seongwook Hwang

Gen AI Engineer · Founder · U.S. Green Card Holder
Gen AI Engineer who
Builds AI Services
Currently a Gen AI Engineer at Samsung SDS America. I founded an AI SaaS (ZipMenu) with multimodal LLMs in production, built enterprise MLOps platforms at MakinaRocks for industrial AI deployment, and have shipped 8+ production AI services including RAG pipelines, LLM agents, and inference infrastructure. Built for scale. Shipped to production.
Gen AI Engineer · Samsung SDS America AI SaaS Founder · ZipMenu MakinaRocks · Enterprise MLOps 6x NVIDIA/AWS/CKA Certs 8+ AI Services Shipped
NVIDIA AI Stack
Certified Professional: AI Operations
Certified Associate: AI Infrastructure and Operations
2x NVIDIA Certs
AWS Certified
ML Engineer · Solution Architect
Data Engineer
3x AWS Certs
🚀
AI SaaS Founder
ZipMenu: Multimodal LLM
+ Instacart Live Orders
zipmenu.info ↗
CKA Certified
Certified Kubernetes
Administrator
Samsung SDS AMERICA
Gen AI Engineer
LangGraph+VectorDB+FASTAPI+Streamlit
Apr 2026 – Present
Active
+25%
Production UPH
Hyundai Transys - MetaPlant
Industrial AI Vision Ops
1M+
Daily Active Users
dktechin - Kakao Music
High-Traffic Backend
Samsung MIRACOM
8 yrs
SDI · SDS · Electronics Project
Mission-Critical Systems
CEO Award Recipient
🇺🇸
U.S. Green Card
No Sponsorship Needed
OHDSI
Int'l Collaborator
OHDSI 2019 Symposium
Medical AI / AWS CDM
ohdsi.org ↗
Yonsei & Hanyang
M.S. Information Systems
B.S. Industrial Engineering
🇰🇷
Republic of Korea Army
Honorable Discharge
Full-Term Service Completed
AI Startup · MLOps Engineer · Mar 2023 – Sep 2024
MakinaRocks
Built Runway: an Enterprise MLOps Platform for industrial AI deployment, serving clients like Corning with Rockwell Automation.
▸ Runway MLOps Platform: data pipeline, model registry, inference placement, SSO JWT auth
▸ ETL pipeline optimization: manufacturing equipment data → Parquet; 25% efficiency gain
▸ Container Image Snapshot & image annotation labeling visualization
▸ Scrum Master: led sprint planning, daily standups, and retrospectives; drove team velocity and delivery cadence
▸ CI/CD pipeline and actively participated in PR reviews: enforced code quality and accelerated release cycles
▸ Led feature prioritization by synthesizing inputs from customers, developers, UI/UX, frontend, and company roadmap, translated into actionable backlog and sprint goals
FastAPI Kubernetes SQL Server Docker MLOps Parquet JWT / SSO Scrum Master CI/CD PR Review
CB Insights AI 100 2024
CB Insights AI 100, 2023

Portfolio

Ordered to tell one story: every RAG/LLM system below is built on MLOps, real-time data pipeline, and large-scale backend foundations I laid over 15+ years at MakinaRocks, Hyundai, Samsung, Kakao, and Rakuten.

🧠 Gen AI & RAG Engineering: Featured Projects
RAG · Compensation Analytics
Market Compensation Insight
A compensation-benchmarking platform that blends public salary data (H1B DB, Zippia) with internal uploads into PostgreSQL + ChromaDB, then serves a React analytics dashboard and a RAG chatbot for natural-language salary queries.
▸ Aggregate-aware RAG: deterministic SQL medians (percentile_cont) fused with top-k vector matches before the GPT-4o-mini call for grounded answers
▸ Dual crawlers: h1bdata.info HTML scraping + Zippia JSON API with autocomplete keyword sweep, run as FastAPI background jobs with status polling
▸ Role × Level × YOE matrix: System / Network / Infra / Helpdesk filled from /salaries/aggregate
▸ Interactive US map: d3-geo + us-atlas markers with click-to-filter by city
▸ Idempotent ingest: source-stable external_id dedup + /ingest/reembed vector rebuild
▸ Automated weekly report: n8n workflow pulls weekly collection counts per source, authenticates via corporate SSO, and delivers the summary through the internal Knox API
RAG VectorDB FastAPI PostgreSQL ChromaDB GPT-4o-mini React Streamlit n8n SSO Knox API
Market Compensation Insight
RAG · Competitive Intelligence
Review Radar
A RAG-powered competitive intelligence dashboard that analyzes thousands of TV product reviews (Samsung vs. LG vs. Sony) into executive-ready insights. Reviews are crawled, chunked, embedded into ChromaDB, and analyzed through 7 structured GPT-4o tasks.
▸ Multi-source crawling: Samsung BFD, LG BFD, and Sony PowerReviews APIs via Playwright interception + httpx
▸ 7-task analysis pipeline: complaints, improvements, marketing, competitive positioning, rating-sentiment mismatch, frequency × severity
▸ Multi-query retrieval: 12 parallel category queries for balanced coverage
▸ SSE streaming chatbot with intent classification, cited evidence reviews, and suggested follow-ups
RAG VectorDB GPT-4o ChromaDB FastAPI React Playwright SSE Streaming
Review Radar: RAG Insight Dashboard
Multi-Model Vision · Risk Scanning
RightsLens
An image copyright & portrait-rights risk scanner that calls OpenAI, Claude, and Gemini Vision in parallel, forces all three into one JSON schema, and cross-checks their findings. Deployed serverless on AWS Lambda with a React frontend.
▸ 3-model cross-check: findings confirmed by ≥2 providers drive a model-agreement score
▸ 5-category detection: famous IP, brand logos, artistic style, copyright notices, portrait rights
▸ Weighted risk scoring: category weight × severity multiplier → 0–100 (Safe / Caution / High Risk)
▸ Forced structured output: OpenAI json_schema, Anthropic tool_use, Gemini response_schema
▸ Graceful degradation: survives partial provider failure via ThreadPoolExecutor
OpenAI Vision Claude Vision Gemini Vision AWS Lambda React
RightsLens
Privacy · Secure External Internet Egress
PII Masking
A privacy that detects and masks PII before internal files leave for external Internet. routing every send through approval. Two-stage detection: Microsoft Presidio rules/NER followed by an Claude verification pass feeds format-preserving masking with full round-trip restore.
▸ 2-stage detection: Presidio (spaCy NER + custom Korean RRN/phone/biz-no & US telecom/device-brand recognizers) → Claude verifier with a degraded fallback
▸ Format-preserving encryption: FF3-1 FPE keeps digit/email layout (010-XXXX-XXXX); a digit-sweep masks every numeric run Presidio/Claude miss
▸ Reversible round-trip: masked → original mappings stored only in HashiCorp Vault; .xlsx cell-level mask & restore with cascade-safe substitution
▸ Audit without exposure: daily ElasticSearch indices log events with no raw PII retained
Claude Model Presidio FastAPI Vault ElasticSearch React
PII Masking
Founder & Solo Engineer · AI Food SaaS
ZipMenu
An AI-powered food recommendation SaaS that turns local weather data and US Zip Codes into personalized dining suggestions, answering the daily "What should I eat?" dilemma. Architected, built, and shipped solo, end to end, from Foundation Model integration to go-to-market.
▸ Multimodal LLM recommendations: weather + zip code context fused into a single prompt for location- and season-aware food suggestions
▸ Instacart API integration: one-click grocery ordering, closing the loop from AI recommendation to real purchase fulfillment
▸ AWS Serverless architecture: full MLOps lifecycle from model integration through deployment, run solo without a dedicated ops team
▸ Stripe billing: automated subscription and payment flow for a live production SaaS
Multimodal LLM Instacart API Live Orders AWS Serverless Stripe
ZipMenu
AI Resume Optimization · ATS
TailorCV
AI-powered resume optimization tool that tailors your resume to pass ATS screening and align with any job description, built with GPT 4.1 and Claude API.
▸ JD Keyword Analysis: match scoring between resume and JD with gap highlights
▸ Resume Reconstruction: LLM rewrites resume content to reflect JD language and priorities
▸ Cover Letter Generator: tailored cover letter with company research and role-specific positioning
▸ Strength Analysis: identifies which experiences are the strongest signals for the target role
▸ Inline Edit + PDF Export: edit directly in-app and export finalized resume and cover letter
React 19 TypeScript GPT 4.1 Claude API n8n PDF Export
Click to watch demo
TailorCV
RAG · Grounded Enterprise Q&A
RAG QA System
A production Retrieval-Augmented Generation system for IT service operations knowledge, built on the exact stack I run at Samsung SDS America: LangGraph · FastAPI · ChromaDB · Streamlit. A stateful LangGraph pipeline retrieves, reranks, and gates every answer against its sources.
▸ 3-layer hallucination guard: confidence gate → citation enforcement → post-generation grounding check
▸ Cross-encoder reranking (ms-marco MiniLM) running locally for top-K → top-N precision
▸ Query rewriting + retry on a failed confidence gate via the LangGraph state machine
▸ Source page rendering: only the PDF pages actually cited are shown to the user
▸ Structured JSON logging with per-node latency (retrieve / rerank / generate)
RAG LangGraph VectorDB FastAPI ChromaDB Streamlit GPT-4o-mini
RAG QA System
Gen AI · Personal Investment Analytics
Stock Analyzer
A personal U.S. stock analysis tool with two views: a sector dashboard that auto-tracks the top names across all 11 GICS sectors with rule-based scoring + Claude AI summaries, and a product component analysis view that decomposes a product (humanoid robot, NVIDIA AI server stack) into its supplier tree to surface the "hidden core" companies behind the headline stock.
▸ 3-backend role separation: n8n (weekly FMP ingest → rule scoring + Claude summary → DynamoDB) as producer, AWS Lambda Function URL as the read API (/analysis, /watchlist), and a FastAPI service for on-demand real-time quotes
▸ Supplier component trees: each product broken into 8 component groups; per-supplier ticker, price, P/E, and a valuation badge (undervalued/fair/overvalued), sorted undervalued-first
▸ ⭐ Hidden-core filter: flags overlooked component suppliers vs. overheated flagship stocks, toggle to isolate them
▸ Global ticker coverage: mixes U.S. and overseas listings (Japan, Korea, Taiwan) for components monopolized abroad, like reducers and HBM
▸ Automated weekly pipeline: n8n cron collects and scores ~22 S&P 500 stocks every Monday without manual intervention
Claude API n8n AWS Lambda DynamoDB FastAPI React TypeScript Cloudflare Pages
Stock Analyzer (finmate)
Cloud Architecture · Multi-AZ Deployment
Chatbot Service on AWS
An AWS reference architecture for a chatbot service serving pharma clients: a Multi-AZ VPC with an ALB in front, auto-scaled EC2 node groups running the chatbot workloads, and PostgreSQL isolated in its own database subnet.
▸ 3-tier network isolation: public subnet (ALB, NAT), private subnet (compute), and a dedicated database subnet for PostgreSQL
▸ Auto Scaling: Master and Worker EC2 node groups in an ASG, load-balanced by the ALB
▸ Chat history storage: SQLite on each node for chat history, with PostgreSQL as the shared persistent store
▸ Controlled egress: external API calls leave only through the NAT Gateway and Internet Gateway
▸ Private service access: VPC Endpoint to ECR and CloudWatch for image pulls and logging, with IAM-scoped permissions
AWS VPC ALB EC2 Auto Scaling PostgreSQL NAT Gateway VPC Endpoint ECR CloudWatch IAM
⚙️ MLOps, Industrial AI & Backend Foundations: The Infrastructure Gen AI Runs On
Runway
MLOps Platform: Runway

Core contributor to Runway, MakinaRocks' enterprise MLOps platform. Engineered end-to-end data pipeline management, model deployment infrastructure, and high-performance backend services on FastAPI, Kubernetes, and SQL Server.

  • Built ETL pipeline transforming manufacturing equipment data into optimized Parquet format
  • Implemented versioned dataset storage for reproducible model training
  • Designed container image snapshot functionality for environment packaging
  • Created image annotation visualization tooling for data labeling workflows
  • Integrated SSO/JWT authentication for enterprise-grade access control
FastAPI Kubernetes SQL Server MLOps
Gen AI today: the same pipeline orchestration, model registry, and deployment discipline now shapes how I ingest, version, and serve RAG corpora and LLM endpoints.
Hyundai
Hyundai Transys: Industrial AI Ops

Deployed at Hyundai Motor Group Metaplant America (HMGMA), Georgia, a $7.6B flagship smart factory and the most automated automotive plant in North America.

  • Integrated AI vision systems with the production floor for real-time defect detection and closed-loop MES feedback
  • Optimized DB performance via partitioning, schema tuning, and connection pooling, reducing CPU load by 7%
  • Refactored MES–PLC communication logic, boosting production throughput (UPH) from 36 to 45 (+25%)
  • Built bidirectional data recording for ADSEM Bolt tool, achieving 100% quality traceability
  • Created shipment and production status dashboard for real-time monitoring and predictive logistics
AI Vision MES PLC Data Pipeline
Gen AI today: wiring AI model output into real-time, closed-loop production workflows is the same discipline behind deploying grounded, low-latency RAG systems that enterprises can actually trust.
Samsung Group
Samsung Group: Enterprise Systems

8 years of mission-critical MES and enterprise systems across Samsung's semiconductor, battery, and solar manufacturing lines. Built production-grade backends in Java, C# .NET, and Oracle with a focus on real-time equipment control and data synchronization.

  • Samsung Electronics FDC: Kafka-based real-time fault detection pipeline; CEO Award recipient
  • Samsung SDI AceLine MES: C# .NET + Oracle + Miracom Solution for battery production line
  • Samsung Fire & Marine: WebLogic Middleware integration and maintenance
  • Samsung SDS SR Portal: Java + MyBatis + Oracle service request portal improvement
  • Hanwha Q CELLS (Dalton, GA): MES initial build for U.S. solar manufacturing plant; C# .NET + Oracle + Miracom Solution
Java C# .NET MES Kafka Oracle Samsung SW Cert. Advance
Gen AI today: Kafka-based real-time fault detection across 8 years of mission-critical systems is the same event-driven pipeline pattern I now use for streaming ingestion into vector stores and RAG corpora.
Kakao
Kakao Music: 1M+ DAU Scale

Optimized backend services for Kakao Music to sustain high-capacity streaming traffic at scale. Built RESTful APIs on Spring Boot with Kafka event streaming, MongoDB persistence, and Kubernetes orchestration, delivering measurable improvements in throughput and system response times.

SpringBoot Kafka MongoDB Kubernetes Batch Processing
Gen AI today: designing RESTful APIs that hold up at 1M+ DAU is the same serving discipline behind running LLM inference and RAG endpoints reliably in production.
Rakuten
Rakuten: Send Anywhere Service

Engineered AWS cloud infrastructure for Send Anywhere's global-scale file transfer service, supporting millions of concurrent users. Built and maintained RESTful APIs with automated CI/CD pipelines (Jenkins, Kubernetes), ensuring high availability and rapid release cycles.

Python Django AWS Jenkins Redis ELK DynamoDB
Gen AI today: global-scale AWS infrastructure and CI/CD are the operational backbone I now rely on to deploy and scale AI/LLM services with the same reliability.
OHDSI
OHDSI: Medical Data Infrastructure

Sole IT engineer in a cross-functional medical research team led by Prof. Rae Woong Park, a globally recognized OHDSI leader.

  • Architected cloud-based clinical data platform on AWS (S3, Redshift), standardizing large-scale EMR datasets into OMOP CDM format
  • Deployed ATLAS research platform on AWS, enabling physicians to run cohort analyses without on-premises constraints
  • Recognized as an official OHDSI International Collaborator at the 2019 Symposium for contributions to medical informatics
AWS Redshift OMOP CDM ATLAS Health Informatics
Gen AI today: standardizing messy, heterogeneous EMR data into one clean schema (OMOP CDM) is the same data-normalization discipline behind building trustworthy retrieval corpora for RAG.

Skills

Software Engineer Stack
  • Model Deployment Automation & GPU Resource Optimization
  • Kubernetes (CKA) Orchestration for Inference Workloads
  • Gen AI & LLM: LangGraph, RAG, VectorDB, Foundation Models, Prompt Engineering, Streamlit
  • MLOps Platform: Data Pipeline, Dataset Versioning, Model Registry, Inference Server, CI/CD
  • Cloud & Infrastructure: AWS (S3, Redshift, EC2, Lambda, DynamoDB), Serverless Architecture
  • Backend Engineering: FastAPI, Spring Boot, C# .NET, RESTful API Design, Microservices
  • Data Engineering: Kafka, MongoDB, SQL Server, Oracle, Redis, ELK, Parquet, ETL Pipelines
  • Industrial AI: MES, PLC Communication, AI Vision Systems, Real-Time Defect Detection
  • Security: CISSP, SSO/JWT Authentication, OAuth
  • Agile & Leadership: Scrum Master, Sprint Planning, Feature Prioritization, PR Reviews

Certifications

  • NVIDIA Certified Professional: AI Operations
  • NVIDIA Certified Associate: AI Infrastructure & Operations
  • CKA: Certified Kubernetes Administrator
  • AWS Certified Machine Learning Engineer Associate
  • AWS Certified Data Engineer Associate
  • AWS Certified Solutions Architect Associate
  • CISSP: Certified Information Systems Security Professional
  • SCJP: SunMicroSystem Certified Java SE Programmer
  • CCNA: Cisco Certified Network Associate Routing and Switching
  • Engineer of Auto Information Processing
  • Korean Hanja Proficiency Certificate Level 2
  • HSK Level 3 - Chinese Proficiency

Experience

Gen AI Engineer

Samsung SDS America, New Jersey, United States

Developing and deploying Generative AI solutions for enterprise clients as part of Samsung SDS America's AI practice. Leveraging expertise in Foundation Models, MLOps pipelines, and cloud infrastructure to deliver production-ready Gen AI systems at scale.

Apr 2026 - Present

Founder & Lead Engineer

Zipmenu

Founded and independently built Zipmenu from the ground up, sole developer responsible for full-stack architecture, product development, and go-to-market launch. Architected a production-ready AI SaaS with full MLOps lifecycle, from Foundation Model integration and AWS Serverless architecture to automated billing via Stripe API. Extended the platform with Instacart API integration, enabling users to place real grocery orders directly from AI-generated food recommendations.

Jan 2026 - Present

Assistant Manager

Hyundai Transys, Hyundai Motor Group Metaplant America (HMGMA), Georgia

Hyundai Motor Group's $7.6B flagship smart factory in the U.S., the most automated automotive plant in North America. Engineered high-speed communication protocols integrating AI vision systems with the production floor, enabling real-time defect detection and closed-loop MES feedback. Optimized large-scale database performance through table partitioning, schema tuning, and connection pooling, reducing server CPU load by 7%. Refactored SQL queries and redesigned indexing strategies to ensure long-term data integrity for quality traceability.

May 2025 - Dec 2025

MLOps Engineer

MakinaRocks

Industrial AI startup recognized as a CB Insights AI 100 (2023) honoree and World Economic Forum Technology Pioneer, backed by $25M+ from Hyundai Motor Group, SK Telecom, GS, and Hanwha. Engineered core features for Runway, MakinaRocks' enterprise MLOps platform, including data pipeline management, dataset versioning, and model deployment infrastructure. Built high-performance backend services using FastAPI and Kubernetes to support AI model inference at enterprise scale. Developed image annotation visualization tooling and SSO/JWT authentication, improving platform usability for enterprise clients.

Mar 2023 - Sep 2024

Cloud Infrastructure Architect & Developer

Rakuten Symphony

Engineered AWS cloud infrastructure supporting large-scale file transfer services for Send Anywhere, handling millions of concurrent users. Built and maintained RESTful APIs with automated CI/CD pipelines (Jenkins, Kubernetes), improving deployment reliability and release velocity.

Feb 2022 - Feb 2023

Backend Engineer, High-Scale Distributed Systems

dktechin (Kakao Group)

Kakao's development subsidiary. Developed and optimized backend services for Kakao Music, handling high-capacity traffic at scale. Built RESTful APIs using Spring Boot, Kafka, and MongoDB to support large-scale streaming services. Implemented Kubernetes-based infrastructure to manage distributed workloads and improve system response times.

Mar 2020 - Feb 2022

Tech Team Lead, Data Infrastructure

Ajou University Medical Center, OHDSI International Collaborator

Led large-scale medical data infrastructure for the OHDSI International Symposium on distributed health research networks. Architected AWS infrastructure (Redshift, S3, EC2) for medical big data research pipelines. Managed CI/CD workflows using Jenkins and Docker, ensuring data governance and pipeline reliability. Recognized as an official OHDSI Collaborator for contributions to the international OMOP CDM research network.

Jul 2019 - Mar 2020

Senior MES Systems Architect

MIRACOM (Samsung Group)

Delivered MES solutions for Samsung Electronics, Samsung SDI, Samsung SDS, and Hanwha Q CELLS America, spanning semiconductor, battery, and solar manufacturing lines. Built backend systems using Java/Spring Boot, C# .NET, and Oracle, with a focus on equipment control and real-time data synchronization. Achieved Samsung SW Certification (Advanced level).

Oct 2011 - Jul 2019

Contract Associate

KOTRA (Korea Trade-Investment Promotion Agency)

Supported foreign direct investment (FDI) promotion initiatives within KOTRA's Foreign Investment Support Division. Aggregated trade office intelligence from global KOTRA branches and produced periodic newsletters distributed to investors and stakeholders. Managed and maintained the Foreign Investment Support web portal, ensuring timely updates and content accuracy. Coordinated logistics and operational support for FDI-related trade fairs and promotional events.

Jan 2010 - May 2011

Retail Banking Associate

Jeju Bank (Shinhan Financial Group), Jeju, South Korea

Jeju Bank, a regional banking institution and affiliate of Shinhan Financial Group. Selected through a competitive nationwide recruitment process. Managed retail banking operations at the branch counter, including customer transactions, account services, and financial product consultation.

Jun 2009 - Dec 2009

Software Engineer, Financial Systems

Lotte Capital (Lotte Group)

Selected through Lotte Group's competitive nationwide recruitment program. Planned and implemented IT strategies for financial services; led Business Process Re-engineering (BPR) and IT Strategic Planning (ISP) initiatives to improve operational efficiency; developed Java-based financial applications with Oracle database integration.

Jan 2008 - May 2009

Education

Yonsei University (Seoul)

Master of Science
Information Systems – Architecture, Design, NLP

Hanyang University (Seoul)

Bachelor of Science
Industrial Engineering – Operations Research

Paper

A Study on the Influence of Happiness Score on the Movement of Residence in the Noise Complaints Focused on the Complaint Board in Seoul

Yonsei University (2019.08)

Key Contribution: Built an automated data pipeline to crawl civil complaint boards in Seoul. Utilized NLP (Natural Language Processing) to analyze social sentiment and proposed data-driven solutions for urban noise issues across Seoul districts.

A comparative Study on the Service Acceptance of Service Users with or without Bandwagon Effects: Focused on Kakao Bank

Presented at APDSI (Asia Pacific Decision Sciences Institute) (2019.07)

Key Contribution: Analyzed the diffusion of fintech services using the 'Bandwagon Effect' theory. Presented research on factors influencing service acceptance among Kakao Bank users at the international APDSI conference.

A Case Study of the Mobile Giving Platforms Based on Construal Level Theory: Focused on Big-walk and Tree Planet

ISR (Information Systems Review) (2015.12)

Key Contribution: Conducted an in-depth case study on social contribution platforms. Recognized by ISR for applying 'Construal Level Theory' to analyze user engagement and business model sustainability in the mobile ecosystem.

Honors & Awards

  • Miracom CEO Award (2020)
    Awarded by the Miracom CEO for the successful delivery of the Hanwha Q CELLS project in the United States.
  • OHDSI International Collaborator (2019)
    Recognized as an international collaborator in the OHDSI (Observational Health Data Sciences and Informatics) global research community.
  • ROK Commander Commendation (2002)
    Awarded twice by a Republic of Korea military commander for distinguished service.