Why Kimchi?

Kimchi is the AI platform inside CAST AI. We started by helping companies run LLMs on their own Kubernetes clusters - now we're building the execution layer where agents do real work.

Our Infrastructure today: multi-model inference (MiniMax, Kimi, GLM-5, Nemotron, DeepSeek) with intelligent routing, an OpenAI-compatible API, and deployment flexibility from our GPUs to your VPC. The inference layer is the foundation. What we're hiring for sits on top of it: coding agents, agent runtimes, orchestration systems, and the reliability engineering that makes them actually finish things.

Tech Stack: TypeScript, Go, Kubernetes, AWS/GCP/Azure, MCP, Prometheus/Grafana/Loki, GitLab CI, ArgoCD.

Why harness engineering matters here
OpenAI and Anthropic ship models. They also ship one harness each - the scaffolding that turns a raw model into something that can plan, execute, recover, and complete work. We ship a different kind of harness: one built for cost-conscious, long-horizon autonomy, running on inference infrastructure we control end-to-end.
A decent model with a great harness beats a great model with a bad harness. We've watched this play out. The gap between what today's models can do and what you see them doing is largely a harness gap - and that gap is where we operate.

What you'll build
The ratchet.
Every time our agent makes a mistake, we engineer a solution so it never makes that mistake again. That means hooks that enforce constraints the model "knows" but forgets: pre-commit lint checks, permission gates, context compaction before the window fills. Success is silent, failures are verbose.

Long-horizon execution.
Our harness is built around spec-driven autonomy: meta-prompting, fresh context per task, worktree-per-slice git strategy, automatic replanning, crash recovery, stuck detection. We're implementing Ralph loops - when the model tries to exit, we intercept and reinject the goal into a fresh context. The agent reads state from disk and continues. Multi-session, multi-day work, without context rot.

Planner/executor splits.
Planning with a reasoning model, executing with a fast one, evaluating with a third. Separating generation from evaluation beats self-verification because agents reliably skew positive when grading their own work.

The harness surface.
CLI, TUI, MCP integration, sandboxed execution, telemetry. Our AGENTS.md is short - every line traces to a specific thing that went wrong. TypeScript on the surface, Go where it matters.

Memory and context.
Moving agents off laptops, giving them state that survives across sessions, managing context so information lands where it's actionable. Compaction, tool-call offloading, progressive skill disclosure.

What makes this different (with receipts)
You've seen the pitch: "we route to the best model." Everyone says that. Here's what we actually have:

• GPU infrastructure we own. Not just an API reseller. From GPU placement across clouds to the inference endpoint your agent calls - we control the cost curve.
• A harness-first thesis. We treat agent failures as configuration problems, not model problems. When we moved from a stock harness to our own, completion rates on internal benchmarks improved by 40%+ on the same model.
• Agents.md that earns every line. No brainstormed rules - every constraint in our system prompt traces to a real failure we saw and fixed.

Requirements:

• You've used AI coding agents in anger. Not demos - real work. You have opinions about Claude Code, Codex, OpenCode, Cursor. You know what it feels like when an agent gets stuck and why.
• Strong TypeScript or Go in production. Comfort moving between them. Our surface is TypeScript; our core is Go.
• You think in harness terms. You read "the agent hallucinated" and your first instinct is to ask what context it was missing, what hook should have caught it, what constraint should exist.
• You drive features end-to-end. Design → build → ship → measure → iterate. We don't have layers that absorb ambiguity for you.

Responsibilities:

• Build and evolve the agent harness - ship hooks, permission gates, and context compaction. Every AGENTS.md constraint traces to a failure you personally diagnosed.
• Own long-horizon execution - multi-session task completion via spec-driven prompting, worktree-per-slice git, Ralph loop recovery, and stuck detection. Completion rate is your metric.
• Architect planner/executor/evaluator pipelines - planning with a reasoning model, execution with a fast one, evaluation with a third. No self-verification.
• Manage agent memory and context - state persistence across sessions, context compaction, tool-call offloading. Zero context rot on multi-day work.
• Own the harness surface - CLI, TUI, MCP integrations, sandboxed execution, telemetry. TypeScript on the surface, Go where it matters.

What success looks like (after 6 months):

• You've shipped at least one major harness feature end-to-end: designed it, built it, measured it, iterated.
• You've added constraints to our AGENTS.md based on failures you personally observed and diagnosed.
• You've improved a measurable reliability metric - completion rate, context efficiency, or cost per successful task.
• You've formed strong opinions about where our harness is load-bearing and where it's dead weight.

What’s in it for you?

• Competitive salary (€6,500 - €9,000 gross, depending on the level of experience).
• Enjoy a flexible, remote-first global environment.
• Collaborate with a global team of cloud experts and innovators, passionate about pushing the boundaries of Kubernetes technology
• Equity options.
• Get quick feedback with a fast-paced workflow. Most feature projects are completed in 1 to 4 weeks.
• Spend 10% of your work time on personal projects or self-improvement. 
• Learning budget for professional and personal development - including access to international conferences and courses that elevate your skills.
• Annual hackathon to spark new ideas and strengthen team bonds.
• Team-building budget and company events to connect with your colleagues.
• Equipment budget to ensure you have everything you need.
• Extra days off to help maintain a healthy work-life balance.

This is a location-specific opportunity. We are currently accepting applications from candidates residing in the following European countries: Bulgaria, Croatia, Estonia, Greece, Hungary, Latvia, Lithuania, Poland, Romania, Slovakia, Slovenia, and Ukraine.

*As part of our standard hiring process, we would like to inform you that a background check may be conducted at the final stage of recruitment through our third-party provider, Checkr.
*Please note that Cast AI does not provide any form of visa sponsorship/work permit.

#LI-Remote

About Shelf

The enterprise is going agentic — but most AI agents fail when they hit real business complexity. Shelf is changing that.

We’ve built the operating system for agentic AI: a platform that models your policies, workflows, and operational logic into an AI Data Model so agents don’t just respond — they reason. The result? AI that understands how your business actually runs and delivers precise, compliant, auditable outcomes at scale.

Brands like Amazon, Nespresso, HelloFresh, and KeyBank trust Shelf to power AI agents that resolve 85% of cases autonomously, cut handle times by 20–25%, and turn hours-long processes into seconds. We’re partnered with Microsoft, Salesforce, OpenAI, Snowflake, and Databricks — and recognized by Gartner (Cool Vendor) and IDC (Innovator) for our approach.

If you want to sell the infrastructure that makes agentic AI actually work in the enterprise, you’re in the right place.

Our mission is to empower humanity with better answers everywhere.

Summary

The R&D department plays a pivotal role in driving Shelf to disrupt the market. We are looking for Machine Learning experts that are able to deliver end to end with a blend of experience: Python engineering, ML engineering, and pragmatic Data science and Machine learning research. You will ship end-to-end features—from problem framing and experimentation to service deployment, and ongoing operations—quickly and with high quality. Your work will power ML- and LLM-driven services used by top enterprises like Amazon, Mayo Clinic, AmFam, and Nespresso.

This role requires strong Python engineering capabilities coupled with a strong ability to deliver robust ML solutions, along with ML research literacy to choose sound methodologies, define metrics, and evaluate different approaches effectively.

You’ll work in an agile environment, move fast, and own what you ship.

Responsibilities

• Own end-to-end delivery: ideate, research, prototype, productionize, and operate ML-powered services with an expectation to iterate and ship frequently
• Stand up robust training/evaluation pipelines: dataset curation, labeling/feedback loops, experiment tracking, offline/online metrics, and A/B testing
• Solve problems using sound methodology, evaluate approaches along with 
• Transform ML models and LLM workflows (including RAG) into reusable, versioned, observable production services with CI/CD
• Collaborate with Product Owners to shape our product and requirements
• Conduct and receive code reviews; champion engineering excellence, testing discipline, and documentation
• Leverage AI coding assistants to accelerate development and create internal agents that automate parts of the engineering workflow
• Share learnings through demos, docs, and knowledge sessions; contribute to a culture of continuous improvement

Requirements

• 3+ years of professional experience researching and shipping ML-based solutions, with strong Python skills and a track record of delivering fast without sacrificing quality
• Proven experience in owning research problems end-to-end, starting from initial data analysis, through iterative research phases to delivering on production
• Practical NLP/LLM experience: transformers, embeddings, prompt design, and evaluation; ability to choose and justify metrics and methodologies
• Strong backend fundamentals: designing RESTful services, schema design, data modeling, and performance tuning for SQL and NoSQL stores
• Data processing skills: pandas/NumPy; experience with batch/stream processing and ETL orchestration (e.g., Airflow, Step Functions)
• Strong English verbal and written communication

As a plus

• LLM ops and safety: eval frameworks (e.g., RAGAS), guardrails, red-teaming, prompt optimization at scale
• Model optimization: quantization, distillation, pruning; GPU/accelerator-aware serving
• Experience with AWS ML stack (SageMaker, Batch, Step Functions, Lambda, SQS/SNS, DynamoDB, ECS, EC2, S3)
• Vector databases and search: Pinecone, Elasticsearch, pgvector, FAISS, or DeepLake
• Background in reinforcement learning, agent frameworks, or autonomous agents
• Publications, open-source contributions, GitHub portfolio

What Shelf Offers

• B2B contract
• Company Stock Options
• Hardware: MacBook Pro
• Modern technical stack. Develop open-source software
• Premier AI development environment: GitHub Copilot, Claude Code, OpenAI, TypingMind, v0, MCP Servers, plus credits to experiment with emerging AI tools

Why Shelf

• Our Leadership Team has deep knowledge management and AI domain expertise and enterprise SaaS background to execute this plan
• We love our customers and our customers love us. Ask a Shelf customer why, and they’ll tell you it’s because of our innovative capabilities, rock-solid reliability, they truly enjoy working with our people, but most of all – it’s the improvements they see in their business KPIs.
• We have raised over $60 million in funding and our investors include Tiger Global, Insight Partners, Connecticut Innovations, and others
• We have high velocity growth powered by the most innovative product in our category, 3X growth for 3 years in a row
• We now have over 100 employees in multiple U.S. states and European countries, and we have ambitious hiring goals over the next few months

Our Values:

Quality - We’re united by our focus on world‑class Quality. Quality in all things – starting with everything that leaves your desk. Everything you touch – every email, report, campaign, and piece of code – should be outstanding. Your work product should blow people away. Having people look at what you’ve done and say, “Wow.” That’s the standard here. Remember that how you do anything is how you do everything. Focus on craftsmanship—your ability to make things better.

Momentum - for us means that you should know that the things you’re responsible for are moving forward. When you look around and see something that’s stalled, get it moving again. We pride ourselves on “ball movement.” When your boss or team leaves you with something, they should return to see measurable progress. Small, continuous movement is our recipe for success. Constantly look for how to make the work around you move forward. We want you to initiate solutions, ideas, and progress. Don’t wait for it to come to you—reach out and create movement. All the time.

Accountability - We expect every team member to feel that they are accountable for more than anyone might normally expect. Each of us should feel real responsibility for things even at the edge of our control. We consistently share and align on expectations, give each other open and respectful feedback, and use those two drivers to ensure that every agreement we make with one another is clear and complete.

Hard Work - We’re here to do something difficult together. We care intensely about the mission and we expect that from our teammates. That care means that we work hard here. Hard work comes with long hours, extra effort…and real opportunity at Shelf. Your passion for creating and sustaining output is a part of our DNA. Support each other, cheer each other on, drive the mission forward. Great teams sustain intense effort together to win.

Learning Agility - We’re innovating in one of the fastest‑moving spaces in history at a time of accelerating global change. That’s incredibly exciting and requires each of us to commit fully to learning each and every day so that we can be the best at what we do. None of us know everything. All of us can learn anything. Staying open and constantly curious is a key success driver at Shelf. It also requires humility. We prize people who are consistently humble and open to making mistakes and growing from them. Recognize also that learning itself is a skill…we need you to be really good at it. Keep dialing in your own understanding about how you learn best and push yourself to keep growing.

Adapt and Thrive - Overcoming challenges lives deep in our DNA. We have a proud history of understanding and living the reality that obstacles are our opportunities…they’re the key to our success. Change is a constant in our business and fighting change is counterproductive. We need you to be good at being uncomfortable and understand that discomfort is the key to growth. Cultivate your own ability to adapt and know that struggling well is something you’ll share with every team you’re on at Shelf. Our company stories are about thriving through real difficulty…together.

Win Together - We win or lose as a team. Always. Everything you do here is connected to the rest of the organization. Part of our shared team environment demands full honesty…real candor and directness with one another. We expect you to constantly be thinking about how to support your teammates and the company, always acting in service to our shared mission and what’s best for the organization as a whole.