AI Product Engineer

Making the worldmake sense.

I want a day with nothing wasted in it — spent only on the work that has to be done and the work you want to do.

whoami

NameJaewoong Ahn
Born1999.02.03
Based inGwangjin-gu, Seoul, South Korea

Career

  • AbearAI Product EngineerMay 2025 Present1 yr 3 mos

    Running the Pervis marketing-AI business and shipping features on Windly.

    Pervis

    • Drove the work from opportunity discovery through to a shipped product
    • Designed and built the AI agent itself as a finite state machine
    • Built auth, the ad platform API integrations, and the performance-data analytics
    • Built the harness engineering system behind AI-driven development
    • Stood up the observability stack — Sentry, Axiom, Langfuse — to trace what the agent does at runtime

    Windly

    • Built Winby, the in-product guidance chatbot
    • Built v2 of the background removal service (BiRefNet-based model)
  • AndreiaBackend & ML Engineer, TPMMay 2023 – Sep 20235 mos

    Led the product engineering team for Ssokdak-Ssokdak, a digital therapeutic that treats early-stage depression through emotion journaling.

    May–Aug 2023 part-time · Aug–Sep 2023 full-time

    • Led the product engineering team and ran project management
    • Built server features for emotion journaling, onboarding, and notifications
    • Extracted emotional signals from journal entries with an LLM
    • Built the recommender that composes demographics, those signals, and the extracted keywords into factors, and matches a user with someone to talk to
    • Designed the service architecture and set up the infrastructure
  • AbearBackend Engineer, InternJan 2023 – Aug 20238 mos

    Shipped features inside Windly, a B2B SaaS platform for e-commerce sellers.

    Jan–Feb 2023 full-time · Mar–Aug 2023 part-time

    • Built v1 of the product category recommender (FastText classifier)
    • Built high-quality background removal (SAM segmentation, ONNX for a lighter model)
    • Built the partner settlement back office
    • Ran the PoC for a payment gateway migration and shipped it
    • Split and tuned databases by data and transaction type
    • Set up backend and frontend CI/CD for the dev environment
    • Introduced product monitoring on Datadog
  • TULONCo-founder, Full-stack EngineerJan 2022 – Sep 20229 mos

    Co-founded TULON, a secure data access platform for machine learning researchers, and owned the backend design and the engineering behind it.

    • Designed a domain-driven architecture and moved the system to microservices
    • Built the back office for data access policy and approvals
    • Built the request-and-approval workflow, the policy management, and the audit log
    • Built the isolated training environment and its data egress controls
    • Built the monitoring system covering both the main service and the isolated training environment

Education

KAISTKorea Advanced Institute of Science and TechnologyB.S.Mar 2018 – Feb 2026Graduated
Major
School of Electrical Engineering
Double major
School of Computing
Minor
Department of Mathematical Sciences
Sejong Academy of Science and ArtsMar 2015 – Feb 2018Graduated

Projects

Stack & languages

Programming LanguagesTypeScript, Python, JavaScript, Go, C/C++
ML / DLPyTorch, Langchain, ONNX, BentoML, Streamlit
WebReact, Next.js, HTML5, CSS
ServerNest.js, FastAPI, Django, Fiber, Flask, Express
Cross Platform AppFlutter, React Native
DatabasePostgreSQL, Supabase, MySQL, MongoDB, DynamoDB, Firebase
Cloud ProvidersAWS, GCP
ObservabilitySentry, Langfuse, Datadog, Axiom, New Relic
DevOps & InfraDocker, Github Actions, Jenkins, RabbitMQ
CollaborationGithub, Slack, Notion, Jira, Confluence
Spoken LanguagesKorean (Native), English (Business proficiency), Chinese (Conversational)

Awards

TecChat — an LLM chatbot for field technical manuals

  • Printed field manuals barely circulate and rarely get used — the service replaced them with a chatbot built on the digitised manuals, in an app of its own
  • Built the workflow that turned the manuals into data
  • Researched a multimodal LLM agent
  • Built the prototype
  • Advanced to the national K-Startup competition (Ministry of National Defense track)

MyVoice — blocking voice-model training with adversarial audio

  • Built a system that poisons AI training with adversarial audio
  • Picked VTubers as the entry market, where the voice is the asset
  • Owned the research into adversarial audio generation and ran the PoC

TurninPoint — hiring developers on portfolio work rather than tests

  • Coding tests and take-home assignments say little about real ability
  • Proposed hiring built around company-specific project work instead
  • Built the MVP full-stack and drove product direction from user interviews

Military service

Battalion S1 Human Resource Specialist (42A)

  • Handled personnel records and administrative support for a battalion of 400+

Senior KATUSA

  • Trained, led, and managed the KATUSA soldiers assigned to the battalion