Research Engineer, AI
for section headings (e.g. About the role, Responsibilities, Requirements, Qualifications, Skills, Benefits, Pay, Schedule). So we should map content accordingly.
We have:
- About Atlys (mission, etc.) Could be "About the role"? Not exactly. But we can treat as "About the role" maybe includes description of role? Actually there is a "The Role" paragraph after "About Atlys". That describes the role. So we could have "About the role" containing that paragraph.
- Responsibilities: The bullet list under "The Job" (multiple sentences). Those are responsibilities.
- Requirements: The "Ideal Candidate" bullet list (skills/experience). Could be Requirements.
- Qualifications: maybe overlapping with Requirements. But we can separate: Qualifications could be education? Not present. Could combine.
- Skills: maybe list of technical skills (Python, PyTorch, etc.) Could be separate.
- Benefits: The "What You Can Expect Atlys" list includes compensation, equity, health insurance, leave policy, meals, learning credits, relocation support, tools, office amenities. That's benefits.
- Pay: Not explicitly stated; only "Competitive compensation and equity". That's vague; we should not create Pay section unless posting actually states them. It says "Competitive compensation and equity". That's not a specific number; it's vague. According to instruction: never emit a heading followed by filler/placeholder text (e.g., "commensurate with experience", "not specified", "N/A", "varies", "TBD"). "Competitive compensation" is similar to vague; we should not create Pay section.
- Schedule: Not mentioned.
Thus we will include sections: About the role, Responsibilities, Requirements (or Ideal Candidate), Skills, Benefits. Possibly Qualifications if we want to separate education but none.
We need to preserve all substantive information; do not invent or summarize away detail.
We need to use - for any list of duties/requirements/benefits, and
for prose.
We need to avoid repeating job title as heading; we don't have job title.
We need to drop boilerplate like cookie/navigation/legal/EEO etc. Not present.
We need to output only HTML fragment.
Let's extract content.
About Atlys paragraph:
"About Atlys Our mission is to enable every person on Earth to travel freely. Travel should feel exciting. Instead, too much of it is still defined by paperwork, uncertainty, and processes that were never designed for travellers. Atlys is changing that. We're building an AI-first platform that removes friction and uncertainty from travel. To date, we've processed visas for 1M+ travellers across 127 countries. But we're still early."
We can put that in
under
About the role
maybe.
But "About the role" maybe should contain description of role: "The Role We're looking for a Research Engineer, AI to build the models and systems that make visa processing faster, more accurate and more reliable than ever before. You'll take ideas from research to production, testing new approaches across LLMs, agents and document intelligence, and shipping the ones that move real outcomes like approval rates, turnaround times and cost per application. This role is for someone with a strong engineering foundation who loves going deep on hard AI problems, and who gets as much satisfaction from a system that works in production as from a clever idea."
We could combine both About Atlys and The Role under About the role? The instruction says use heading for section headings e.g. About the role. So we can put the role description there. The company mission could be maybe also under About the role or we could add a separate "About the company" but not in list. Since only allowed headings are those, we could include company mission in About the role as well, but that might be mixing. However we need to preserve all substantive information; we can include both paragraphs under About the role.
Thus About the role
then containing both paragraphs (maybe two
tags). But we can combine into one
with line breaks? Better to separate into two
for readability.
Now Responsibilities: The Job paragraph: "Build AI systems that do real work by designing and shipping LLM-powered agents that reliably complete complex, multi-step tasks across external tools, APIs and web interfaces.Push the frontier of document intelligence by building pipelines that extract, verify and reason over information from unstructured documents and images with high accuracy at scale.Define how we measure AI quality by building evaluation frameworks, benchmarks and datasets that show how accurate, reliable and safe our models are across the long tail of real-world cases, before and after they ship.Run experiments that change decisions by comparing models, prompting strategies, fine-tuning approaches and routing logic to improve accuracy while reducing latency and cost.Own systems end to end by taking work from prototype to production, including data pipelines, inference infrastructure, monitoring and human-in-the-loop fallbacks for when models aren’t confident."
These are sentences; we can treat each as a list item (
- ) under
. The instruction: Use - for any list of duties/requirements/benefits. So we will make a
with each sentence as - . Ensure we preserve all detail.
Now Requirements: The Ideal Candidate bullet list: "You'll Likely Thrive In This Role If You Have 3-6 years of software engineering experience, including meaningful work building and shipping ML or LLM-based systems in production.Are exceptionally strong in Python and comfortable with the modern AI stack, such as PyTorch, Hugging Face, LLM APIs, vector databases and agent frameworks.Have hands-on experience with LLMs in practice, including prompting, RAG, fine-tuning, tool use and evals, and understand where each approach works and where it breaks.Bring a research mindset with a bias to ship: you form hypotheses, run clean experiments, read papers when they help, and turn results into systems people use.Have exposure to document AI, OCR, large-scale data pipelines, or making inference fast and cost-efficient at scale."
These are sentences; we can make
each as - . Could be under Requirements heading.
Now Skills: maybe we can extract specific technical skills: Python, PyTorch, Hugging Face, LLM APIs, vector databases, agent frameworks, prompting, RAG, fine-tuning, tool use, evals, document AI, OCR, large-scale data pipelines, inference optimization. But these are already in Requirements. However we could still create a Skills section listing them. But we must not duplicate content? The instruction says preserve ALL substantive information; does not forbid duplication but better to avoid redundancy. However we could create Skills section summarizing the technical skills from the Ideal Candidate. But we need to ensure we don't invent or summarize away detail. If we list them, we are extracting from the text, not inventing. That's okay.
But we need to decide: Should we have both Requirements and Skills? The instruction says prefer a few substantive sections over many thin ones. So maybe combine Requirements and Skills into one section (Requirements) and not create Skills. However the instruction list includes Skills as possible heading; we can include if there is real specific content. There is specific content about skills (Python, PyTorch, etc.) but they are already in the Ideal Candidate bullet list. If we make a separate Skills section, we would be repeating content. Might be okay but could be considered thin. Better to not create Skills section; just have Requirements covering both experience and skills.
Thus sections: About the role, Responsibilities, Requirements, Benefits.
Now Benefits: The "What You Can Expect Atlys" paragraph: "Competitive compensation and equity so you participate in the value you help create.Comprehensive health insurance covering you and your family, plus annual preventive health check-up on us.A thoughtful leave policy with paid time off and company holidays, because great work also requires real time off.Lunch, high tea and dinner at the office, so the basics are taken care of while we build together.Learning & development credits to invest in courses, books, conferences and tools that help you get better at your craft.Relocation support to make moving closer to the team as seamless as possible.Great tools to do your best work, including high-spec equipment and software.An office designed for more than just work, with pickleball, table tennis, PlayStation, an arcade game room, an F1 simulator and even a full flight simulator."
We need to make each as
- under
. Ensure we preserve all detail.
Now we need to exclude boilerplate like "Our values shape how we make decisions, work together and build for travelers. The bar is high, the pace is fast, and the way we work is deliberate. Read them before you apply. If this environment energizes you, you'll probably thrive here. If it doesn't, it's better to know that upfront. Come build the intelligence layer for global travel". This is somewhat fluff but still substantive? It's about values and encouragement. Might be considered boilerplate? It's not cookie/navigation/legal/EEO etc. It's more about culture. The instruction says remove cookie/navigation/legal/EEO/equal-opportunity/scam-warning boilerplate, duplicated lines, and application instructions. This is not exactly those; it's more about culture. Could keep? But it's somewhat generic. However it's still substantive info about company values. The instruction says prefer a few substantive sections over many thin ones. This could be a thin section; we might omit to keep sections few. But we already have About the role covering mission; we could incorporate values there? Might be better to omit.
Thus final HTML:
About the role
About Atlys Our mission is to enable every person on Earth to travel freely. Travel should feel exciting. Instead, too much of it is still defined by paperwork, uncertainty, and processes that were never designed for travellers. Atlys is changing that. We're building an AI-first platform that removes friction and uncertainty from travel. To date, we've processed visas for 1M+ travellers across 127 countries. But we're still early.
The Role We're looking for a Research Engineer, AI to build the models and systems that make visa processing faster, more accurate and more reliable than ever before. You'll take ideas from research to production, testing new approaches across LLMs, agents and document intelligence, and shipping the ones that move real outcomes like approval rates, turnaround times and cost per application. This role is for someone with a strong engineering foundation who loves going deep on hard AI problems, and who gets as much satisfaction from a system that works in production as from a clever idea.
Responsibilities
- Build AI systems that do real work by designing and shipping LLM-powered agents that reliably complete complex, multi-step tasks across external tools, APIs and web interfaces.
- Push the frontier of document intelligence by building pipelines that extract, verify and reason over information from unstructured documents and images with high accuracy at scale.
- Define how we measure AI quality by building evaluation frameworks, benchmarks and datasets that show how accurate, reliable and safe our models are across the long tail of real-world cases, before and after they ship.
- Run experiments that change decisions by comparing models, prompting strategies, fine-tuning approaches and routing logic to improve accuracy while reducing latency and cost.
- Own systems end to end by taking work from prototype to production, including data pipelines, inference infrastructure, monitoring and human-in-the-loop fallbacks for when models aren’t confident.
Requirements
- You'll Likely Thrive In This Role If You Have 3-6 years of software engineering experience, including meaningful work building
for prose. We need to avoid repeating job title as heading; we don't have job title. We need to drop boilerplate like cookie/navigation/legal/EEO etc. Not present. We need to output only HTML fragment. Let's extract content. About Atlys paragraph: "About Atlys Our mission is to enable every person on Earth to travel freely. Travel should feel exciting. Instead, too much of it is still defined by paperwork, uncertainty, and processes that were never designed for travellers. Atlys is changing that. We're building an AI-first platform that removes friction and uncertainty from travel. To date, we've processed visas for 1M+ travellers across 127 countries. But we're still early." We can put that in
under
About the role
maybe. But "About the role" maybe should contain description of role: "The Role We're looking for a Research Engineer, AI to build the models and systems that make visa processing faster, more accurate and more reliable than ever before. You'll take ideas from research to production, testing new approaches across LLMs, agents and document intelligence, and shipping the ones that move real outcomes like approval rates, turnaround times and cost per application. This role is for someone with a strong engineering foundation who loves going deep on hard AI problems, and who gets as much satisfaction from a system that works in production as from a clever idea." We could combine both About Atlys and The Role under About the role? The instruction says use heading for section headings e.g. About the role. So we can put the role description there. The company mission could be maybe also under About the role or we could add a separate "About the company" but not in list. Since only allowed headings are those, we could include company mission in About the role as well, but that might be mixing. However we need to preserve all substantive information; we can include both paragraphs under About the role. ThusAbout the role
thencontaining both paragraphs (maybe two
tags). But we can combine into one
with line breaks? Better to separate into two
for readability. Now Responsibilities: The Job paragraph: "Build AI systems that do real work by designing and shipping LLM-powered agents that reliably complete complex, multi-step tasks across external tools, APIs and web interfaces.Push the frontier of document intelligence by building pipelines that extract, verify and reason over information from unstructured documents and images with high accuracy at scale.Define how we measure AI quality by building evaluation frameworks, benchmarks and datasets that show how accurate, reliable and safe our models are across the long tail of real-world cases, before and after they ship.Run experiments that change decisions by comparing models, prompting strategies, fine-tuning approaches and routing logic to improve accuracy while reducing latency and cost.Own systems end to end by taking work from prototype to production, including data pipelines, inference infrastructure, monitoring and human-in-the-loop fallbacks for when models aren’t confident." These are sentences; we can treat each as a list item (
- . The instruction: Use
- for any list of duties/requirements/benefits. So we will make a
- with each sentence as
- . Ensure we preserve all detail.
Now Requirements: The Ideal Candidate bullet list: "You'll Likely Thrive In This Role If You Have 3-6 years of software engineering experience, including meaningful work building and shipping ML or LLM-based systems in production.Are exceptionally strong in Python and comfortable with the modern AI stack, such as PyTorch, Hugging Face, LLM APIs, vector databases and agent frameworks.Have hands-on experience with LLMs in practice, including prompting, RAG, fine-tuning, tool use and evals, and understand where each approach works and where it breaks.Bring a research mindset with a bias to ship: you form hypotheses, run clean experiments, read papers when they help, and turn results into systems people use.Have exposure to document AI, OCR, large-scale data pipelines, or making inference fast and cost-efficient at scale."
These are sentences; we can make
- each as
- . Could be under Requirements heading. Now Skills: maybe we can extract specific technical skills: Python, PyTorch, Hugging Face, LLM APIs, vector databases, agent frameworks, prompting, RAG, fine-tuning, tool use, evals, document AI, OCR, large-scale data pipelines, inference optimization. But these are already in Requirements. However we could still create a Skills section listing them. But we must not duplicate content? The instruction says preserve ALL substantive information; does not forbid duplication but better to avoid redundancy. However we could create Skills section summarizing the technical skills from the Ideal Candidate. But we need to ensure we don't invent or summarize away detail. If we list them, we are extracting from the text, not inventing. That's okay. But we need to decide: Should we have both Requirements and Skills? The instruction says prefer a few substantive sections over many thin ones. So maybe combine Requirements and Skills into one section (Requirements) and not create Skills. However the instruction list includes Skills as possible heading; we can include if there is real specific content. There is specific content about skills (Python, PyTorch, etc.) but they are already in the Ideal Candidate bullet list. If we make a separate Skills section, we would be repeating content. Might be okay but could be considered thin. Better to not create Skills section; just have Requirements covering both experience and skills. Thus sections: About the role, Responsibilities, Requirements, Benefits. Now Benefits: The "What You Can Expect Atlys" paragraph: "Competitive compensation and equity so you participate in the value you help create.Comprehensive health insurance covering you and your family, plus annual preventive health check-up on us.A thoughtful leave policy with paid time off and company holidays, because great work also requires real time off.Lunch, high tea and dinner at the office, so the basics are taken care of while we build together.Learning & development credits to invest in courses, books, conferences and tools that help you get better at your craft.Relocation support to make moving closer to the team as seamless as possible.Great tools to do your best work, including high-spec equipment and software.An office designed for more than just work, with pickleball, table tennis, PlayStation, an arcade game room, an F1 simulator and even a full flight simulator." We need to make each as
- under
- . Ensure we preserve all detail.
Now we need to exclude boilerplate like "Our values shape how we make decisions, work together and build for travelers. The bar is high, the pace is fast, and the way we work is deliberate. Read them before you apply. If this environment energizes you, you'll probably thrive here. If it doesn't, it's better to know that upfront. Come build the intelligence layer for global travel". This is somewhat fluff but still substantive? It's about values and encouragement. Might be considered boilerplate? It's not cookie/navigation/legal/EEO etc. It's more about culture. The instruction says remove cookie/navigation/legal/EEO/equal-opportunity/scam-warning boilerplate, duplicated lines, and application instructions. This is not exactly those; it's more about culture. Could keep? But it's somewhat generic. However it's still substantive info about company values. The instruction says prefer a few substantive sections over many thin ones. This could be a thin section; we might omit to keep sections few. But we already have About the role covering mission; we could incorporate values there? Might be better to omit.
Thus final HTML:
- Build AI systems that do real work by designing and shipping LLM-powered agents that reliably complete complex, multi-step tasks across external tools, APIs and web interfaces.
- Push the frontier of document intelligence by building pipelines that extract, verify and reason over information from unstructured documents and images with high accuracy at scale.
- Define how we measure AI quality by building evaluation frameworks, benchmarks and datasets that show how accurate, reliable and safe our models are across the long tail of real-world cases, before and after they ship.
- Run experiments that change decisions by comparing models, prompting strategies, fine-tuning approaches and routing logic to improve accuracy while reducing latency and cost.
- Own systems end to end by taking work from prototype to production, including data pipelines, inference infrastructure, monitoring and human-in-the-loop fallbacks for when models aren’t confident.
- You'll Likely Thrive In This Role If You Have 3-6 years of software engineering experience, including meaningful work building
About the role
About Atlys Our mission is to enable every person on Earth to travel freely. Travel should feel exciting. Instead, too much of it is still defined by paperwork, uncertainty, and processes that were never designed for travellers. Atlys is changing that. We're building an AI-first platform that removes friction and uncertainty from travel. To date, we've processed visas for 1M+ travellers across 127 countries. But we're still early.
The Role We're looking for a Research Engineer, AI to build the models and systems that make visa processing faster, more accurate and more reliable than ever before. You'll take ideas from research to production, testing new approaches across LLMs, agents and document intelligence, and shipping the ones that move real outcomes like approval rates, turnaround times and cost per application. This role is for someone with a strong engineering foundation who loves going deep on hard AI problems, and who gets as much satisfaction from a system that works in production as from a clever idea.
Responsibilities
Requirements
- . Ensure we preserve all detail.
Now Requirements: The Ideal Candidate bullet list: "You'll Likely Thrive In This Role If You Have 3-6 years of software engineering experience, including meaningful work building and shipping ML or LLM-based systems in production.Are exceptionally strong in Python and comfortable with the modern AI stack, such as PyTorch, Hugging Face, LLM APIs, vector databases and agent frameworks.Have hands-on experience with LLMs in practice, including prompting, RAG, fine-tuning, tool use and evals, and understand where each approach works and where it breaks.Bring a research mindset with a bias to ship: you form hypotheses, run clean experiments, read papers when they help, and turn results into systems people use.Have exposure to document AI, OCR, large-scale data pipelines, or making inference fast and cost-efficient at scale."
These are sentences; we can make