aimlroles

AI Engineer (Brain, AIOC)

Toss Open Β· verified Oct 22, 2026
Seoul Mid-level AI Engineering

At a glance

Mid-level AI Engineering role at Toss. Seoul.

Pay not stated AI Engineer salary

RoleAI Engineering
SeniorityMid-level
LocationSeoul
PostedMar 2, 2026 Β· 7mo ago
Role brief

aimlroles summary, based on the employer's posting.

What you'll do

  • Build core capabilities that let text, image and real-time voice agents understand customers and carry out business tasks
  • Experiment with prompts, RAG, context design, evaluations, tool use and safeguards to make agents effective and safe
  • Create measurement and validation pipelines that reveal weaknesses in uncertain AI components and support repeated improvement
  • Use customer feedback and business metrics to refine agent behavior, including multimodal understanding and low-latency voice interaction
  • Design solutions that balance technical quality with practical deployment and long-term viability

What you bring

  • You have used Multi-Agent systems, LLMs, RAG or multimodal models to address complex business problems
  • You have built and repeatedly improved production AI agents, including text- or image-based systems
  • You can assess and tune trade-offs among quality, response time and cost
  • You have worked with varied data such as text, images, audio or structured records to develop and test models

Who this fits

This role suits an engineer who can take AI work from problem definition through experiments and quantitative evaluation, then improve it using customer and business evidence. Voice experience is welcome but not essential if you have production agent experience with text or images. The process includes a technical interview covering in-depth AI topics and ML systems design.

From the employer

πŸ“μ§λ¬΄ μΈν„°λ·°λ§Œ 봐도 100λ§Œμ›, ν•¨κ»˜ν•  νŒ€μ„ 직접 λ§Œλ‚˜λ³΄κ³  κ²°μ •ν•˜μ„Έμš”!
- 10μ›” 8일(λͺ©)λΆ€ν„° 25일(금)에 μ§€μ›ν•˜μ‹œλŠ” λΆ„λ“€μ—κ²Œλ§Œ ν•΄λ‹Ήλ˜λŠ” νŠΉλ³„ μ „ν˜•μ΄μ—μš”. 합격 여뢀와 상관없이 직무 인터뷰에 μ°Έμ—¬ν•˜μ‹œλ©΄ 100λ§Œμ›μ„ λ°›μ•„μš”. (μ œμ„Έκ³΅κ³ΌκΈˆ 본인 λΆ€λ‹΄)
- 총 7개의 직ꡰ, 40μ—¬κ°œμ˜ ν¬μ§€μ…˜μ—μ„œ μ±„μš©μ„ μ§„ν–‰ν•˜κ³  μžˆμ–΄μš”. (https://toss.im/career/event/ai-ml-data-2026)
- μ§€μ›ν•˜μ‹  ν¬μ§€μ…˜λ³΄λ‹€ 더 잘 λ§žλŠ” ν¬μ§€μ…˜μ΄ μžˆλ‹€λ©΄ 저희가 λ¨Όμ € μ œμ•ˆλ“œλ €μš”. λ‹€λ₯Έ 직무둜 λ³€κ²½ν•˜μ—¬ 인터뷰λ₯Ό μ§„ν–‰ν•˜μ…”λ„ 100λ§Œμ›μ€ λ™μΌν•˜κ²Œ μ§€κΈ‰λΌμš”.

# ν•©λ₯˜ν•˜κ²Œ 될 νŒ€μ— λŒ€ν•΄ μ•Œλ €λ“œλ €μš”
- 이 ν¬μ§€μ…˜μ€ ν† μŠ€μ—μ„œ AIκ°€ ν…μŠ€νŠΈΒ·μ΄λ―Έμ§€ 기반 μ—μ΄μ „νŠΈμ™€ μ‹€μ‹œκ°„ μŒμ„± μ—μ΄μ „νŠΈλ₯Ό 톡해 고객과 μžμ—°μŠ€λŸ½κ²Œ λŒ€ν™”ν•˜κ³ , μ‹€μ œ λΉ„μ¦ˆλ‹ˆμŠ€λ₯Ό μˆ˜ν–‰ν•˜λŠ” μ‹œμŠ€ν…œμ„ λ§Œλ“œλŠ” νŒ€μ— ν•©λ₯˜ν•΄μš”.
- ν…μŠ€νŠΈΒ·μ΄λ―Έμ§€ 기반 μ—μ΄μ „νŠΈλŠ” 고객의 μ˜λ„μ™€ λ§₯락을 μ΄ν•΄ν•˜κ³ , RAGλ₯Ό 톡해 ν•„μš”ν•œ 정보λ₯Ό μ°Ύμ•„μ˜€λ©°, Tool Calling으둜 μ μ ˆν•œ μ„œλΉ„μŠ€ κΈ°λŠ₯을 μ‹€ν–‰ν•΄ 고객의 문제λ₯Ό ν•΄κ²°ν•΄μ•Ό ν•΄μš”.
- μ‹€μ‹œκ°„ μŒμ„± μ—μ΄μ „νŠΈλŠ” 고객이 말을 λŠκΈ°λ„ ν•˜κ³ , λ§₯락을 κ°‘μžκΈ° 바꾸기도 ν•˜κ³ , 침묡이 흐λ₯΄κΈ°λ„ ν•˜λŠ” μƒν™©μ—μ„œλ„ μžμ—°μŠ€λŸ½κ²Œ λŒ€ν™”λ₯Ό 이어가야 ν•΄μš”. Out-bound 콜의 ν˜Ήλ…ν•œ ν™˜κ²½μ—μ„œλ„ λͺ©μ μ„ 달성할 수 μžˆλŠ” 지λŠ₯적인 μ„œλΉ„μŠ€λ₯Ό λ§Œλ“€κ³  μžˆμ–΄μš”.
- λ‹¨μˆœν•œ λŒ€ν™”λ₯Ό λ„˜μ–΄μ„œ, μ‹€μ œ λΉ„μ¦ˆλ‹ˆμŠ€λ₯Ό μˆ˜ν–‰ν•˜λŠ” AI μ—μ΄μ „νŠΈλ₯Ό λ§Œλ“€μ–΄μš”. 고객을 움직일 수 μžˆλŠ” λ˜‘λ˜‘ν•œ AIλ₯Ό μœ„ν•΄ 각 λΆ„μ•Όμ˜ μ „λ¬Έκ°€κ°€ νž˜μ„ λͺ¨μ•„ μ›νŒ€μœΌλ‘œ μΌν•΄μš”.
- μ΅œμ‹  AI κΈ°μˆ μ„ μ‹€μ œ 고객과의 μƒν˜Έμž‘μš©μ΄λΌλŠ” κ°€μž₯ κΉŒλ‹€λ‘œμš΄ ν™˜κ²½μ— μ μš©ν•˜κ³ , 맀일 고객 ν”Όλ“œλ°±κ³Ό λ°μ΄ν„°λ‘œ κ²€μ¦ν•΄μš”.
- λ©€ν‹°λͺ¨λ‹¬ 이해와 LLM을 ν™œμš©ν•œ μ‹€μ‹œκ°„ μŒμ„± λŒ€ν™”λŠ” 아직 업계에도 정닡이 μ—†λŠ” μ˜μ—­μ΄μ—μš”. 논문을 μ½λŠ” κ²ƒμ—μ„œ λ©ˆμΆ”μ§€ μ•Šκ³ , μ‹€μ œ 고객 λ°μ΄ν„°λ‘œ 직접 닡을 λ§Œλ“€μ–΄κ°€λŠ” νŒ€μ΄μ—μš”.

# ν•©λ₯˜ν•˜λ©΄ ν•¨κ»˜ ν•  μ—…λ¬΄μ—μš”
- ν…μŠ€νŠΈΒ·μ΄λ―Έμ§€ 기반 μ—μ΄μ „νŠΈμ™€ μ‹€μ‹œκ°„ μŒμ„± μ—μ΄μ „νŠΈμ˜ λ‘λ‡Œ 역할을 ν•˜λŠ” 핡심 κΈ°λŠ₯듀을 μ„€κ³„ν•˜κ³  κ΅¬ν˜„ν•΄μš”.
- ν”„λ‘¬ν”„νŠΈ, μ—μ΄μ „νŠΈ ꡬ성, RAG, Context Engineering, Evaluation, Tool Calling, Guardrail λ“± AI μ—μ΄μ „νŠΈκ°€ 각 ν™˜κ²½μ—μ„œ 효과적이고 μ•ˆμ „ν•˜κ²Œ λͺ©μ μ„ 달성할 수 μžˆλ„λ‘ μ—¬λŸ¬ μž₯치λ₯Ό κ³ μ•ˆν•˜κ³  μ‹€ν—˜ν•΄μš”.
- λΆˆν™•μ‹€ν•˜κ³  ν™•λ₯ μ μΈ AI/ML μ»΄ν¬λ„ŒνŠΈλ₯Ό μ›ν•˜λŠ” λͺ©μ μ— 맞게 μ•ˆμ „ν•˜κ³  효과적으둜 μ‚¬μš©ν•  수 μžˆλ„λ‘ μ •λŸ‰ν™”λœ 평가와 검증을 μˆ˜ν–‰ν•˜κ³ , 반볡적으둜 κ°œμ„ λ˜λŠ” νŒŒμ΄ν”„λΌμΈμ„ μ„€κ³„ν•΄μš”.
- μ‹€μ œ 고객 ν”Όλ“œλ°±κ³Ό λΉ„μ¦ˆλ‹ˆμŠ€ μ§€ν‘œμ—μ„œ κ°œμ„ μ μ„ μ°Ύκ³  반볡적으둜 κ°œμ„ ν•˜μ—¬, 고객과 μžμ—°μŠ€λŸ½κ²Œ μƒν˜Έμž‘μš©ν•˜κ³  λΉ„μ¦ˆλ‹ˆμŠ€ μ„±κ³Όλ₯Ό λ§Œλ“€μ–΄λ‚΄λŠ” AI μ—μ΄μ „νŠΈλ₯Ό λ§Œλ“€μ–΄μš”.
- ν…μŠ€νŠΈΒ·μ΄λ―Έμ§€ 기반 μ—μ΄μ „νŠΈμ—μ„œλŠ” λ‹€μ–‘ν•œ λ©€ν‹°λͺ¨λ‹¬ μž…λ ₯κ³Ό ν’λΆ€ν•œ λ§₯락을 μ •ν™•ν•˜κ²Œ μ΄ν•΄ν•˜κ³  ν™œμš©ν•  수 μžˆλ„λ‘ ν•˜λ©°, μ‹€μ‹œκ°„ μŒμ„± ν™˜κ²½μ—μ„œλŠ” μ €μ§€μ—°Β·κ³ ν’ˆμ§ˆμ˜ λŒ€ν™” κ²½ν—˜μ„ λ§Œλ“€ 수 μžˆλ„λ‘ 지λŠ₯적이고 효율적인 μ—μ΄μ „νŠΈ ꡬ성을 μ„€κ³„ν•΄μš”.
- 기술적 정ꡐ함뿐 μ•„λ‹ˆλΌ λΉ„μ¦ˆλ‹ˆμŠ€ 적용 κ°€λŠ₯μ„±κ³Ό 지속 κ°€λŠ₯μ„±κΉŒμ§€ κ³ λ €ν•œ μ†”λ£¨μ…˜μ„ μ„€κ³„ν•΄μš”. μ§€κΈˆκΉŒμ§€ λͺ…ν™•νžˆ μ •μ˜λ˜μ§€ μ•Šμ•˜λ˜ μ˜μ—­μ— AIλ₯Ό μ μš©ν•  수 μžˆλŠ” κ°€λŠ₯성도 νƒμƒ‰ν•΄μš”.

# 이런 λΆ„κ³Ό ν•¨κ»˜ν•˜κ³  μ‹Άμ–΄μš”
- 문제 μ •μ˜λΆ€ν„° λͺ¨λΈ 섀계, μ‹€ν—˜, μ •λŸ‰ ν‰κ°€κΉŒμ§€ 전체 과정을 μ£Όλ„ν•œ κ²½ν—˜μ΄ 있으면 μ’‹μ•„μš”. λ°μ΄ν„°μ…‹Β·μ§€ν‘œΒ·κ°€λ“œλ ˆμΌμ„ 기반으둜 ν’ˆμ§ˆμ„ μ²΄κ³„μ μœΌλ‘œ λŒμ–΄μ˜¬λ¦΄ 수 μžˆλŠ” 뢄을 μ°Ύκ³  μžˆμ–΄μš”. λΆˆν™•μ‹€μ„±μ„ μ •λŸ‰ν™”ν•˜λŠ” 데 λŠ₯μˆ™ν•˜κ³ , 본인만의 검증 방식을 κ°–μΆ˜ 뢄이면 μ’‹μ•„μš”.
- Multi-Agent, LLM, RAG, λ©€ν‹°λͺ¨λ‹¬ λͺ¨λΈ λ“± μ΅œμ‹  AI κΈ°μˆ μ„ ν™œμš©ν•΄ λ³΅μž‘ν•œ λΉ„μ¦ˆλ‹ˆμŠ€ 문제λ₯Ό ν•΄κ²°ν•œ κ²½ν—˜μ΄ ν•„μš”ν•΄μš”. κΈ°μˆ μ„ μœ„ν•œ 기술이 μ•„λ‹ˆλΌ, 고객 κ²½ν—˜κ³Ό λΉ„μ¦ˆλ‹ˆμŠ€ μ„±κ³Όλ₯Ό κ·ΉλŒ€ν™”ν•˜κΈ° μœ„ν•œ 고민이 μ€‘μš”ν•΄μš”.
- 기술적 μ‹€ν—˜λΏ μ•„λ‹ˆλΌ μ„œλΉ„μŠ€ 적용 κ°€λŠ₯μ„±κ³Ό ν™•μž₯μ„±κΉŒμ§€ κ³ λ €ν•œ 섀계 κ²½ν—˜μ΄ 있으면 μ’‹μ•„μš”.
- ν’ˆμ§ˆΒ·μ§€μ—° μ‹œκ°„Β·λΉ„μš© μ‚¬μ΄μ˜ νŠΈλ ˆμ΄λ“œμ˜€ν”„λ₯Ό μ΄ν•΄ν•˜κ³ , 상황에 λ§žλŠ” 선택과 μ΅œμ ν™”λ₯Ό ν•΄λ³Έ 뢄이면 μ’‹μ•„μš”.
- ν…μŠ€νŠΈ, 이미지, μŒμ„±, ꡬ쑰화 데이터 λ“± λ‹€μ–‘ν•œ ν˜•νƒœμ˜ 데이터λ₯Ό ν†΅ν•©μ μœΌλ‘œ ν™œμš©ν•΄ λͺ¨λΈμ„ μ„€κ³„ν•˜κ³  μ‹€ν—˜ν•΄λ³Έ 뢄이 ν•„μš”ν•΄μš”.
- μŒμ„± κ΄€λ ¨ κ²½ν—˜μ΄ 없어도, ν…μŠ€νŠΈΒ·μ΄λ―Έμ§€ κΈ°λ°˜μ„ ν¬ν•¨ν•œ ν”„λ‘œλ•μ…˜ 레벨의 AI μ—μ΄μ „νŠΈλ₯Ό κ°œλ°œν•˜κ³  반볡적으둜 κ°œμ„ ν•œ κ²½ν—˜μ΄ μžˆλ‹€λ©΄ μΆ©λΆ„ν•΄μš”.

# 이λ ₯μ„œλŠ” μ΄λ ‡κ²Œ μž‘μ„±ν•˜μ‹œλŠ” κ±Έ μΆ”μ²œν•΄μš”
- λ‹¨μˆœν•œ λͺ¨λΈλ§ 기술 λ‚˜μ—΄λ³΄λ‹€λŠ”, 무엇을 κ°œμ„ ν•˜κ³  μ–΄λ–€ 영ν–₯을 μ£Όμ—ˆλŠ”μ§€λ₯Ό ꡬ체적으둜 λ³΄μ—¬μ£Όμ‹œλ©΄ μ’‹μ•„μš”.
- κ²½ν—˜μ— λŒ€ν•œ κ³΅μœ λŠ”, μ œμ•½ 상황 β†’ 문제 μ •μ˜ β†’ μ‹€ν—˜ 섀계 β†’ 평가 β†’ κ°œμ„ μ˜ 과정이 잘 λ“œλŸ¬λ‚˜λ„λ‘ 정리해 μ£Όμ‹œλ©΄ μ’‹μ•„μš”.
- ν˜‘μ—… κ³Όμ •μ—μ„œ μ–΄λ–€ 역할을 ν–ˆλŠ”μ§€, 기술 μ™Έμ μœΌλ‘œλ„ 문제 해결에 μ–΄λ–»κ²Œ κΈ°μ—¬ν–ˆλŠ”μ§€ λ“œλŸ¬λ‚˜λ©΄ 더 μ’‹μŠ΅λ‹ˆλ‹€.

# ν† μŠ€λ‘œμ˜ ν•©λ₯˜μ—¬μ •
- μ„œλ₯˜ μ ‘μˆ˜ > 직무 인터뷰 > 문화적합성 인터뷰 > 레퍼런슀 체크 > 처우 ν˜‘μ˜ > μ΅œμ’… 합격
- 직무 μΈν„°λ·°μ—μ„œλŠ” 심측 기술 λ©΄μ ‘κ³Ό ML μ‹œμŠ€ν…œ 섀계λ₯Ό 주제둜 면접이 진행될 μ˜ˆμ •μ΄μ—μš”.

# ν•¨κ»˜ν•  λ™λ£Œλ₯Ό μœ„ν•œ ν•œλ§ˆλ””
> β€œλ§ν•˜κ³  λ“£λŠ” AI 기술둜 고객의 κ²½ν—˜μ„ λ°”κΎΈλŠ” μ„œλΉ„μŠ€λ₯Ό λ§Œλ“€κ³  μžˆμ–΄μš”. 세상을 λ°”κΎΈλŠ” AI κΈ°μˆ μ„ ν•¨κ»˜ λ§Œλ“€ λ™λ£Œλ₯Ό μ°ΎμŠ΅λ‹ˆλ‹€!”
> - 세상에 μ—†λ˜ zero to one μ„œλΉ„μŠ€μ΄κΈ° λ•Œλ¬Έμ—, 고객 κ΄€μ μ—μ„œ 문제λ₯Ό μ •μ˜ν•˜κ³  κ°œμ„  방법을 μ°Ύμ•„λ‚΄λŠ” 관점이 μ€‘μš”ν•΄μš”.
> - λ‹¨μˆœ λͺ¨λΈλ§ 반볡 μž‘μ—…μ΄ μ•„λ‹ˆλΌ, AIλ₯Ό λΉ„λ‘―ν•œ λ‹€μ–‘ν•œ 툴 μ…‹μœΌλ‘œ 주어진 문제λ₯Ό μœ μ—°ν•˜κ²Œ ν’€μ–΄λ‚΄μ•Όν•΄μš”.
> - μš°λ¦¬κ°€ λ§Œλ“  AIκ°€ μ„±μž₯ν•˜λŠ” 과정에 직접 κΈ°μ—¬ν•  수 μžˆμ–΄μš”. λ‹¨μˆœνžˆ μ§ˆλ¬Έμ— 잘 λŒ€λ‹΅ν•˜λŠ” 챗봇을 λ„˜μ–΄, λΉ„μ¦ˆλ‹ˆμŠ€ νŒŒνŠΈλ„ˆλ‘œ μΌν•˜λŠ” λ˜‘λ˜‘ν•œ AIλ₯Ό ν•¨κ»˜ λ§Œλ“€μ–΄κ°€μš”.