New opportunity

Applied AI Researcher, System Self-Construction

Distyl AI · San Francisco, United States

About this role

Distyl AI seeks an Applied AI Researcher on its System Self-Construction team to research systems that can build other systems. The team develops AI technologies for enterprise operations, with an emphasis on approaches that can change how software is created and used.

Responsibilities

Design architectures that autonomously generate, assemble, and refine subsystems. Develop meta-learning loops, automated pipeline generation, and self-architecting frameworks to reduce human effort in complex system creation.
Study recursive design: how systems represent, evaluate, and evolve their own construction processes. Apply meta-learning, program synthesis, and evolutionary computation to self-referential, generative architectures.
Build prototypes and run experiments that demonstrate the effectiveness of research ideas for enterprise customers.

Qualifications

Experience engineering modular, composable systems whose components interact dynamically, such as graph-based planners, agent orchestration frameworks, workflow engines, or automated pipeline builders.
Experience building intelligent systems with models rather than primarily training or fine-tuning models. Ideal candidates have expertise in compound AI systems, agentic collaboration, and techniques such as ensembling, ReAct, and graph-of-thoughts.
A demonstrated research track record, strong programming and data analysis skills, and the ability to prototype and test ideas. Candidates should regularly use AI tools such as ChatGPT, Cursor, and Perplexity and favor practical demonstrations of results. Researchers may come from varied academic backgrounds; no specific degree or number of years of experience is stated.

Location and compensation:

The posting lists San Francisco and New York. The role is hybrid, with at least three in-office days per week, Tuesday through Thursday.
Base salary is $150K–$250K, depending on experience, location, and level, with equity eligibility. Benefits include medical, dental, and vision coverage for employees and dependents, flexible time off, retirement and financial-planning resources, wellness and family-building benefits, and in-office meals and snacks.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

Meta-learningProgram synthesisEvolutionary computationModular systemsComposable systemsGraph-based planningAgent orchestrationWorkflow enginesAutomated pipeline generationCompound AI systemsAgentic collaborationEnsemblingReActGraph-of-thoughtsProgrammingData analysisExperimentationPrototypingChatGPTCursorPerplexity

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Autonomous system construction
The team designs architectures that generate, assemble, and refine subsystems to reduce human effort. Prepare to explain an architecture’s construction loop and where human intervention remains necessary.
Recursive design and evaluation
The role studies how systems represent, assess, and evolve their own construction processes, drawing on meta-learning, program synthesis, and evolutionary computation. Prepare to distinguish a genuinely adaptive design loop from a fixed pipeline.
Composable systems built with models
The posting emphasizes dynamically interacting components and intelligent systems built using models, rather than primarily training models. Prepare a concrete account of component boundaries, coordination, and why the chosen composition helps.
Research demonstrated through prototypes
Distyl asks researchers to build prototypes, analyze experiments, and show practical effectiveness to enterprise customers. Prepare to connect a research result to a working demonstration, evidence of improvement, and its limitations.
AI-native research practice
The posting values a visible research track record and regular use of AI tools to accelerate work. Prepare to explain how such tools fit into a research workflow without substituting for independent validation.

Likely challenges to think through

Autonomy versus control
A likely design tension is reducing the human bottleneck while retaining a way to inspect and correct generated subsystems. Prepare to discuss explicit constraints, review points, and failure cases as proposed design choices, not stated company practices.
Evaluating self-improvement
A recursive system could optimize a convenient metric without improving the resulting system. Prepare a hypothetical evaluation plan with baselines, held-out tasks, and checks on both subsystem quality and construction effort.
Novel research versus a useful demonstration
The posting favors creative research but also says customers want to see AI’s power today. Prepare to show how a small, testable prototype could support an ambitious idea while making its current limits clear.

Practice your answers

Technical practice

How would you design a system that generates, assembles, and refines its own subsystems?

What to coverThe role focuses on architectures that reduce human effort in complex system creation. Suggested preparation: explain the components, how they are assembled, and how the system decides whether a refinement is an improvement.

How could a system represent its own construction process?

What to coverThe posting identifies representation, evaluation, and evolution of construction processes as recursive-design problems. Suggested preparation: describe a representation that can be inspected and modified, including what information it must preserve.

What would a meta-learning loop change in a self-constructing system?

What to coverThe team develops meta-learning loops for system construction. Suggested preparation: distinguish learning to improve the construction process from improving one generated subsystem, and explain how feedback would guide the next iteration.

When would you use program synthesis rather than evolutionary computation to generate a subsystem?

What to coverBoth approaches are named as relevant to the team's research. Suggested preparation: compare their search methods, evaluation needs, and suitability when useful structure or constraints are known.

How would you make dynamically interacting components composable?

What to coverThe posting cites graph-based planners, agent orchestration frameworks, and workflow engines as examples of modular systems. Suggested preparation: discuss component interfaces, dependencies, and how you would detect integration failures when components change.

How would you diagnose an automatically generated pipeline that fails on some inputs?

What to coverAutomated pipeline generation is a stated area of work. Suggested preparation: explain how you would trace failures to a component or assembly decision, then test whether a proposed change improves the pipeline without breaking other cases.

How would you choose among ReAct, ensembling, and graph-of-thoughts in a compound AI system?

What to coverThe posting names these techniques and emphasizes building intelligent systems with models rather than primarily training them. Suggested preparation: compare when each approach helps, its coordination costs, and how you would test whether added complexity is worthwhile.

How would you evaluate whether self-construction reduces human effort?

What to coverReducing the human bottleneck is an explicit goal of the team. Suggested preparation: define a baseline, measure both system effectiveness and human work required, and account for cases where generated systems need substantial repair.

What prototype and experiment would demonstrate a research idea's value for an enterprise operation?

What to coverThe posting calls for prototypes, data analysis, and experiments that demonstrate effectiveness to enterprise customers. Suggested preparation: propose a bounded demonstration, a relevant comparison, and evidence that distinguishes a working result from a compelling demo alone.

How would you use AI tools to accelerate research without trusting their output unchecked?

What to coverThe posting expects regular use of tools such as ChatGPT, Cursor, and Perplexity. Suggested preparation: explain which research or prototyping tasks you would delegate, and how you would verify generated code, claims, and experimental conclusions.

Behavioural practice

Tell me about a research result you made visible to others.

What to coverThe posting values a demonstrated research track record across different forms of sharing. Suggested preparation: give a real situation, the research actions you took, the result you shared, and what evidence supported it.

Describe a time an experiment challenged your initial research direction.

What to coverThe role involves evaluating and refining new system architectures. Suggested preparation: use a real situation to explain what the evidence showed, how you changed course, and the resulting outcome.

Tell me about a time you turned an idea into a working prototype quickly.

What to coverThe posting emphasizes showing practical results rather than only discussing long-term ideas. Suggested preparation: describe the situation, what you chose to build or defer, your actions, and what the prototype demonstrated.

Describe a time you explained a technical result to someone focused on operational value.

What to coverThe posting calls for experiments that can demonstrate effectiveness to enterprise decision-makers. Suggested preparation: recount a real situation, how you communicated the evidence and its limits, and the result of that conversation.

Tell me about a time you improved your research or development workflow with AI tools.

What to coverRegular use of AI tools to accelerate work is stated in the posting. Suggested preparation: describe the original bottleneck, your specific actions, how you checked the tool's output, and the observed result.

Sharpen your knowledge

Topics to prepare

Recursive system design
Sketch how a system could represent, evaluate, and revise its own construction process; identify what feedback would justify a redesign. This directly reflects the team's stated research focus.
Meta-learning, program synthesis, and evolutionary computation
Compare how each approach could generate or improve a subsystem, including its search space and evaluation signal. The posting explicitly names all three as foundations for self-construction research.
Modular and composable architectures
Diagram a graph-based planner or workflow engine with dynamically interacting components, then explain how a component could be replaced without redesigning the whole system. The role explicitly seeks experience with such architectures.
Compound AI system design
Contrast a single-model approach with agentic collaboration, ensembling, ReAct, or graph-of-thoughts for the same task; explain the trade-offs rather than assuming greater complexity is better. The posting emphasizes building systems with models, not primarily training them.

Skills to revise

Prototype an automated pipeline
Build a small program that assembles interchangeable steps, runs the resulting pipeline, and refines one step based on its output. This practices the advertised work of generating, assembling, and improving subsystems.
Run and analyze a comparative experiment
Test a prototype against a simpler fixed pipeline on the same tasks; record outcomes, analyze failures, and state the limits of the comparison. Strong programming, data analysis, and experiments that demonstrate effectiveness are explicit expectations.
Demonstrate a research result
Prepare a short walkthrough of a research result: the question, prototype, evidence, and what did not work. The posting asks for a research track record and favors showing practical results to enterprise customers.
Use AI tools in a research workflow
Practice using a tool named in the posting—ChatGPT, Cursor, or Perplexity—to accelerate a prototype or analysis task, then verify its output and note what required correction. The posting explicitly expects regular use of AI tools.

Your preparation checklist

Map the role to preparation materials
The posting focuses on systems that generate, assemble, and refine other systems. Make a one-page map linking those responsibilities to relevant work you can discuss and identifying gaps to explore in a design exercise.
Prepare a modular-system walkthrough
If you have a relevant project, bring an architecture diagram showing how components interact dynamically, such as in a planner, orchestration framework, workflow engine, or pipeline builder. Explain your design choices and what you learned.
Work through a self-construction design exercise
Sketch a system that represents its own pipeline, proposes changes, evaluates them, and retains improvements. Label any use of meta-learning, program synthesis, or evolutionary computation as a proposed approach, not an established Distyl design.
Assemble research and experiment evidence
Select research results you can substantiate and prepare a concise account of the question, prototype, baselines, measurements, analysis, and limitations. The posting emphasizes demonstrable results that can make sense to an enterprise AI leader.
Show how you build with models and AI tools
Prepare an example of a compound-AI workflow using models rather than primarily training them. Explain component coordination and, where relevant, how tools such as ChatGPT, Cursor, or Perplexity accelerated your work and how you checked their output.
Draft questions about research priorities
Ask how the team evaluates reduced human effort in system creation, chooses enterprise problems for prototypes, and decides when a self-constructing approach is effective enough to demonstrate.

Availability last checked: Nov 4, 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in United States. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…