AI · Full-Stack Product Engineering

Hyunseo Oh

I build AI systems from retrieval and evaluation pipelines to user-facing products.

Data Science undergraduate at Sookmyung Women’s University working across LLM/RAG systems, human-in-the-loop workflows, React/FastAPI products, and AI automation.

Selected work GitHub ↗ LinkedIn ↗ Email

Hyunseo Oh presenting her research poster at KDD-UC 2026

Selected work

Projects

01 · Re:mind · 2026

AI documentation workspace for counselors

Live demo ↗ GitHub ↗

Problem

Counselors spend substantial time turning session notes and transcripts into structured records while checking whether AI-generated content is supported by source material.

What I built

React + FastAPI + LangGraph workflow for document processing, prior-session retrieval, structured extraction, draft generation, evidence mapping, validation, and human review.

My role

Team Lead · product specification · AI workflow design · backend/API implementation · LLM/RAG integration · counselor interviews · MVP evaluation.

Input→Document processing→Prior-session retrieval→Structured extraction→Draft generation→Evidence mapping→Validation→Human review

React
TypeScript
FastAPI
LangGraph
Pydantic
Supabase
PostgreSQL
OpenAI API

02 · Reflector · 2025

Retrieval-augmented conversational system

GitHub ↗

Reflector guided reflection chat session

Built a deployed conversational AI pipeline combining dialogue-state estimation, embedding-based retrieval, conditional RAG, response routing, structured validation, session persistence, and longitudinal visualization.

Dialogue-state dependent retrievalEmbedding-based knowledge retrievalConditional RAG & routingStructured response validationPersistent session historyRecharts visualization

React
TypeScript
Supabase
OpenAI API
Recharts
Vercel

03 · DABIDA / SAT Automation

State-synchronized browser automation

Chrome Extension automating multi-stage SAT workflows inside Gemini and exporting structured question, answer, and explanation data into PDFs.

DOM / Shadow DOM extractionAsync UI-state synchronizationRetry & recoveryAutomated navigationPDF generation

Chrome MV3
JavaScript
jsPDF
html2canvas
esbuild

GitHub ↗

04 · UKHA Showroom

Next.js deployed website

Designed and deployed a responsive showroom website with dynamic collection and piece-detail routes, motion, and image-led navigation.

Next.js 16
React 19
Tailwind CSS
Framer Motion
Vercel

Live ↗GitHub ↗

Research

KDD 2026 · Undergraduate Consortium · Accepted

When Retrieval Helps: Selective Retrieval for Single-Turn Mental-Health QA

QLoRA fine-tuning · BM25 retrieval · selective RAG gating · automated LLM evaluation · threshold calibration.

Code & experiments ↗

Selective risk-aware denoising framework diagram

CIKM 2026 · Short Paper Submission

SFT-First Selective Risk-Aware Denoising

LoRA refinement · localized residual-risk detection · span-level repair · response quality and safety evaluation.

Code & experiments ↗

Experience

Jan 2026 — Present · Seoul

Undergraduate Research Assistant

AI Inference Lab · Sookmyung Women’s University

LLM fine-tuning, retrieval, evaluation, adversarial testing, and longitudinal evidence retrieval.

Aug 2025 — Feb 2026 · Pangyo

Front-End / AI Engineering Intern

DABIDA · GENITEACHER

Teacher-facing React/TypeScript product features and AI-assisted workflow automation.

Recognition

Grand Prize at the 2026 SW Startup Idea Competition

Excellence Award at the 2025 Sookmyung AI Hackathon

Excellence Award at the 2025 Industry-Academic Cooperation Project

Excellence AwardStartup Idea Competition · 2026

1st PlaceDaewoong Foundation × Sookmyung AI Hackathon · 2025

Excellence AwardIndustry-Academic Cooperation Project · 2025

Stack

AI / MLPython · PyTorch · Hugging Face · PEFT/LoRA/QLoRA · RAG · BM25 · LLM evaluation

Backend / AgentsFastAPI · LangGraph · Pydantic · Supabase · PostgreSQL · OpenAI API

FrontendReact · TypeScript · Next.js · Vite · Recharts · Vercel