ML Research Intern in Agentic Runtime Systems

About the role

Our mission

Constructor’s mission is to enable all educational organisations to provide high-quality digital education to 10x people with 10x efficiency.

With strong expertise in machine intelligence and data science, Constructor’s all-in-one platform for education and research addresses today’s pressing educational challenges: access inequality, tech clutter, and low engagement of students.

Please send your resume in English only.

Interns are NOT eligible for equipment and other benefits as the position is temporary.

The Reality

The frontier models are converging into commodities; the engine is no longer the differentiator. The scarcity in 2026 is not the model.

We are not looking for people to fine-tune Llama on toy datasets or write "better prompts." We are building the runtime systems that turn probabilistic token generation into reliable workers that finish 50-step tasks without spiraling. We are solving the math of multi-step reliability through engineering, not wishful thinking.

What You Will Solve

You will work on the Agentic Harness, treating the LLM as a component within a larger, deterministic control system. Your research will focus on:

- Architecting loops that survive error accumulation. You will build verification layers that force the agent to recognize its own hallucinations and self-correct.

- You will implement systems where the agent compiles its own constraints and the harness enforces them via constrained decoding or external execution.

- You will optimise KV-cache hit rates and design "restorable compression" strategies to prevent attention decay.

Who We Are Looking For

We are looking for systems thinkers who happen to know ML, not the other way around.

- You know that a demo is not a product. You are interested in the "boring" engineering: state machines, retries, idempotency, serialisation — that actually makes agents work.

- You are comfortable working with the raw output of the model and building robust software in Python or Rust around it.

- You understand why dynamic tool loading destroys inference economics. You prefer deterministic guardrails over "prompting it to be nicer."

- You are willing to read a 500-line trace of a failed agent loop to find the exact moment the context drifted. You care about the "last mile" of reliability.

What We Offer For Interns:

- Internship Duration: from 2 months or more

- Type of internship:

- 100% remote

- or hybrid in the office in Bremen (GER), Sofia (BG), Belgrade (SRB), Istanbul (TR), Sakarya (TR), Singapore (SGD)

What We Offer

- 💻 Choice of work equipment (e.g., laptop, monitor, etc.)

- 🇬🇧 English classes (iTalki – $130 monthly)

- ⏰ Flexible schedule (we usually work between 09:00/10:00 and 18:00/19:00)