2026 Winter Internship (Research)

• About Lablup

  Under the motto "Make AI Accessible," Lablup develops and operates Backend.AI, an innovative AI infrastructure operations platform. From AI model development to training, deployment, and management, we work to minimize the complexity of running AI infrastructure. We help compute workloads of every scale run efficiently, so that developers, researchers, and operators can focus on what matters most.

• Internship Program
  • Share your work and what you've learned through a midterm and final presentation
  • Join "Pebble Seminar," our internal forum on the latest technologies and deep learning research
  • Work alongside lead mentors from each area to broaden your experience and skills

  * Note: this internship does not include or guarantee a full-time offer.

• Responsibilities
  • Track 1. Model Evaluation & Serving Optimization
    • Model evaluation and leaderboard operation: run the full evaluation cycle yourself, from model intake to measurement to reporting
    • Review the validity of scoring criteria (LLM-as-judge rubrics, multiple-choice grading, etc.) and evaluation data
    • Analyze why scores change as measurement conditions change
    • Serving optimization: derive recommended serving settings for each model (quantization, context length, concurrency) based on measurements
  • Track 2. AI Agents & Platform Automation
    • Natural-language platform control agent: design and build an agent that interprets users' natural-language requests to allocate resources, manage sessions, and change settings on an AI infrastructure platform
    • Deliver an end-to-end platform operation experience through a conversational interface
    • Redesign the architecture on a modern agent stack (Backend.AI Go, MLXcel, Continuum Hub, Google ADK, etc.)
    • Cluster operations intelligence agent: design and build an agent that monitors GPU cluster health and predicts and alerts on signs of failure in advance; set up metric- and log-based anomaly detection and alerting
    • Productization: turn internal-tool-level agents into Lablup products (packaging, configuration, documentation, demo scenarios)
• Internship Goals
  • Track 1
    • Run the full evaluation cycle on multiple models and write evaluation reports
    • Review the validity of scoring criteria and evaluation data, and propose improvements
    • Write a guide to recommended serving settings for each model
  • Track 2
    • A working demo of an agent that performs core platform management tasks (session creation, resource allocation, model serving, etc.) through natural language
    • Integrate the cluster monitoring and failure-prediction alert agent with the platform and validate it on real-world scenarios
    • Package both agents as installable, operable products
    • Write user documentation and demo scenarios
• Requirements
  • Common
    • Experience with Linux and Git-based development environments
    • Ability to read technical documents in English and summarize and share them in Korean
    • Interest in AI and software, and the drive to see problems through to the end
  • Track 1
    • Experience with PyTorch and the Hugging Face ecosystem
    • Experience with LLM inference serving (vLLM, SGLang, etc.)
    • Understanding of evaluation metrics and experiment design (able to explain what a metric actually measures; familiar with concepts such as sample size and error)
  • Track 2
    • Experience developing with LLM agent frameworks or tool calling
    • Experience building agentic workflows with LLM platforms and APIs
    • Basic knowledge of REST API design and web service development
    • Experience with container environments (Docker)
• Preferred Qualifications
  • Common
    • Experience using Backend.AI or other MLOps platforms
    • Experience training and serving models in GPU environments
    • Open-source contributions, technical blog posts, or talks
  • Track 1
    • Experience building LLM benchmarks or designing LLM-as-judge
    • Hands-on measurement of quantization (AWQ, GPTQ, FP8, etc.)
    • Experience analyzing evaluation results, such as statistical significance testing
    • Experience working in air-gapped environments
  • Track 2
    • Experience building AI agentic workflows with Backend.AI Go and Token Factory
    • Experience with modern agent stacks such as Google ADK and MCP
    • Experience with time-series metric-based anomaly detection or monitoring systems (Prometheus, Grafana, etc.)
    • Experience developing chat-style UIs
    • Experience writing product documentation
• Working Conditions & Benefits
  • Work at Lablup HQ (1 minute from Seonjeongneung Station)
  • 5 days a week, 8 hours a day (10:00–19:00)
  • One day off per month during the internship
  • Opportunities to join developer communities and events
  • Lunch provided
  • Delicious snacks always available
• Hiring Process
  • Document Screening > Interview > Final Offer
    • Application period: from the posting date until Sunday, November 15, 2026, 23:59 (KST)
    • Interviews: Wednesday, November 18 – Friday, November 27, 2026
    • Final results: expected in the first week of December 2026
  • Internship tracks: Agentic AI / Research Track 1 & 2 / Frontend / GTM
    • You may apply to more than one track. Please rank your choices (up to two) in the application form. Your ranking is used as a reference for placement, and applying with a second choice will not count against you.
  • Please complete the application form via the link below and attach your resume (CV) in any format.
  • The offer may be withdrawn if any submitted information is found to be false or if any reason arises that disqualifies you from employment.