New opportunity

Full-Stack AI Engineer

MarcoPolo Learning · Bengaluru, India

About this role

MarcoPolo Learning seeks a Full-Stack AI Engineer to own the intelligence behind Audrey, its AI Teaching Companion & Coach for early-learning educators. The role spans retrieval, generation, application logic, backend and frontend integration, with substantial influence over the product as it approaches launch and scales.

Responsibilities

Build fast, accurate retrieval that helps teachers find appropriate videos, printables and activities. Ingest and structure videos, guides, printables and multilingual content through content and data pipelines.
Design Audrey’s responses, behavior and personality; improve the quality, speed and reliability of its answers and recommendations. Integrate the AI into the wider product and work with infrastructure engineers on efficient cloud operation.
Establish evaluation, testing, monitoring and clear technical documentation. Collaborate with engineering and product teams, contribute to the roadmap for personalization, multilingual support and classroom insights, and build with privacy and safety in mind, including FERPA/COPPA alignment.

Requirements

5+ years of full-stack software engineering experience and hands-on experience building and shipping at least one production LLM/AI application.
Strong experience with RAG systems, including embeddings, vector databases, semantic search and re-ranking; strong Python, backend fundamentals in APIs, databases and data pipelines, and sufficient frontend ability to work end to end.
Practical LLM orchestration and prompt engineering, including agent and tool design, system prompts, personas and guardrails; experience with LangChain or a comparable tool. Fluency with AI-assisted development tools such as Cursor and solid AWS experience across compute, storage, databases, containers and monitoring.
Product awareness, independent ownership, collaboration and strong written communication while working remotely across UK and US time zones.

Preferred

Familiarity with Amazon Bedrock or self-hosted model inference; edtech or early-childhood education experience; multilingual retrieval; video/audio ML; and familiarity with FERPA/COPPA or similar requirements.

Location and engagement: Fully remote within India, with Bengaluru preferred and meaningful daily overlap with UK and US teams. Applicants must have the right to work in India; visa sponsorship is unavailable. The description calls the engagement full-time, permanent and ongoing, while the careers page classifies its employment type as Contractor. It offers flexible daily scheduling and a competitive, market-aligned salary without a stated amount.

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

Skills for this role

PythonRetrieval-Augmented Generation (RAG)Large Language Models (LLMs)EmbeddingsVector databasesSemantic searchRe-rankingData pipelinesAPIsDatabasesFrontend developmentBackend developmentLLM orchestrationPrompt engineeringLangChainAI agentsTool designSystem-prompt designPersona designAI guardrailsAWSAmazon BedrockSelf-hosted model inferenceCursorAI-assisted developmentCloud monitoringAI evaluationTestingTechnical documentationProduct collaboration

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

Teacher-relevant retrieval
Own fast, accurate retrieval so educators can find appropriate videos, printables, and activities. The role calls for RAG depth in embeddings, semantic search, vector databases, and re-ranking.
Content pipelines for precise search
Ingest and structure videos, guides, printables, and multilingual content so Audrey can retrieve useful materials reliably.
Audrey’s guidance and safety
Shape Audrey’s responses, behavior, and personality for early-learning educators while designing guardrails and building with privacy and safety in mind, including FERPA/COPPA alignment.
Launch-ready, end-to-end AI
Connect retrieval and generation to the backend and frontend, help bring Audrey to launch, and partner with infrastructure engineering on efficient cloud operation.
Measurable quality and product direction
Establish evaluation, testing, monitoring, and clear documentation; improve answer quality, speed, and reliability while collaborating with product and engineering on future capabilities.

Likely challenges to think through

Precision across varied content
Likely preparation challenge, not a reported problem: videos, printables, guides, and multilingual material may need different structuring, while teachers still need fast, relevant results. Prepare to explain how you would assess retrieval quality and choose between richer indexing and response speed.
Improving an AI experience near launch
Likely preparation challenge: refining answers and recommendations while integrating Audrey into the product could put quality, latency, and cloud efficiency in tension. Prepare a method for prioritizing changes and using evaluation and monitoring to check their effects.
Useful guidance within safety boundaries
Likely preparation challenge: more personalized guidance and classroom insights could increase the need to handle educator and child-related information carefully. Prepare to discuss how you would limit data use, test guardrails, and seek appropriate review for FERPA/COPPA alignment without assuming a particular implementation.

Practice your answers

Technical practice

How would you design retrieval for teachers searching for a suitable video, printable or activity?

What to coverSuggested answer focus: Explain how embeddings, vector search, metadata and re-ranking could work together, and how you would handle a request with several teaching constraints. Discuss the trade-off between retrieval accuracy and response speed.

How would you prepare videos, guides and printables for reliable retrieval?

What to coverSuggested answer focus: Walk through ingestion, extracting useful content, structuring metadata and updating an index when materials change. Explain how you would detect missing or poorly structured content before it affects recommendations.

How would you approach retrieval across multilingual content?

What to coverSuggested answer focus: Compare possible indexing and search approaches across languages, including how language-specific content might be matched to a teacher’s request. Describe how you would test relevance by language rather than assuming one approach performs equally well everywhere.

How would you design Audrey’s responses and tool use for an educator-facing assistant?

What to coverSuggested answer focus: Discuss system prompts, persona, when a tool or retrieval step is warranted, and guardrails for unsupported answers. Explain how you would balance a helpful coaching style with accurate, grounded recommendations.

How would you connect the AI experience across the backend and frontend?

What to coverSuggested answer focus: Describe Python API boundaries, data flow and frontend states for a teacher requesting materials or guidance. Cover error handling and how the interface would communicate a slow, unavailable or uncertain response.

How would you evaluate whether Audrey’s answers and recommendations are improving?

What to coverSuggested answer focus: Propose separate checks for retrieval relevance, answer grounding, usefulness and safety, with representative educator requests. Explain how automated tests and human review could inform changes before and after launch.

A previously reliable recommendation becomes slow or inaccurate. How would you investigate?

What to coverSuggested answer focus: Trace the content pipeline, retrieval, re-ranking, generation and API path to isolate the regression. Discuss what monitoring evidence you would inspect and how you would weigh a quick mitigation against a durable fix.

How would you operate an LLM-backed product efficiently on AWS?

What to coverSuggested answer focus: Discuss the roles of compute, storage, databases, containers and monitoring in an architecture you might propose. Explain latency, reliability and cost trade-offs; managed inference such as Amazon Bedrock versus self-hosting is an optional comparison, not a stated requirement.

What privacy and safety checks would you build into classroom-facing AI features?

What to coverSuggested answer focus: Discuss limiting sensitive data exposure, controlling access and testing responses for unsafe or inappropriate output. Explain how you would seek review of FERPA/COPPA alignment rather than claiming an engineering design alone establishes compliance.

How would you use AI-assisted development tools without weakening code quality?

What to coverSuggested answer focus: Describe where a tool such as Cursor might speed up implementation or iteration and where independent review remains necessary. Cover tests, security checks and clear documentation for AI-generated changes.

Behavioural practice

Tell me about a time user needs changed your technical approach.

What to coverSuggested answer focus: Use situation, action and result to show how you learned what users actually needed and adjusted a product decision. If relevant, connect the lesson to understanding educators rather than assuming their workflows.

Describe a project you drove from an ambiguous goal toward launch.

What to coverSuggested answer focus: Set out the situation, the decisions and actions you owned, and the resulting outcome. Show how you prioritized quality and reliability while working with limited oversight, without presenting an invented achievement.

Tell me about a disagreement across product and engineering teams.

What to coverSuggested answer focus: Explain the situation, how you shared evidence and listened to other constraints, and the result. Highlight how you built alignment without overstating consensus or your sole contribution.

Give an example of coordinating work across remote teams or time zones.

What to coverSuggested answer focus: Describe the situation, your written updates or documentation, and how you used limited overlap to resolve decisions. State the result and what you would improve in a similar collaboration.

Describe a time you had to balance a promising feature against privacy, safety or quality risks.

What to coverSuggested answer focus: Use situation, action and result to explain the trade-off, whose input you sought and what you decided. Relate the reasoning to the posting’s planned personalization or classroom insights only as a suggested application, not as an existing feature.

Sharpen your knowledge

Topics to prepare

Retrieval accuracy and ranking
The posting explicitly calls for RAG using embeddings, vector databases, semantic search and re-ranking to help teachers find appropriate materials. Suggested exercise: explain how you would compare initial retrieval with re-ranked results for a teacher seeking a specific activity, including accuracy and speed trade-offs.
Content structure for retrieval
Audrey must ingest videos, printables, guides and multilingual content. Suggested exercise: sketch what information each content type would need for precise retrieval, and explain how missing or inconsistent information could affect recommendations.
LLM behavior and guardrails
The role includes orchestration, agent and tool design, system prompts, personas and guardrails. Suggested exercise: outline how Audrey should respond when a teacher’s request is ambiguous or retrieved materials do not support a confident answer.
Evaluation and operational reliability
The posting requires evaluation, testing and monitoring while improving answer quality, speed and reliability. Suggested exercise: define a small set of teacher queries and describe how you would assess retrieval relevance, answer support and response time before launch.
Privacy and safety in an education product
The responsibilities explicitly include privacy and safety, including FERPA/COPPA alignment; prior familiarity is listed as a bonus. Suggested exercise: identify where teacher or child-related information might enter an AI interaction and discuss what should be reviewed before using it for retrieval or personalization.

Skills to revise

Build a Python content pipeline
Strong Python and data-pipeline skills are required. Suggested exercise: write a small pipeline that ingests sample video descriptions, printable records and guides, normalizes their fields, and reports records that cannot be indexed reliably.
Implement and test RAG retrieval
Hands-on RAG experience is an explicit requirement. Suggested exercise: index a small sample of educator-facing materials in a vector database, run semantic searches, add re-ranking, and inspect which queries still return unsuitable results.
Prototype an orchestrated assistant response
The posting asks for practical LLM orchestration and prompt engineering with LangChain or a comparable tool. Suggested exercise: create a retrieval-backed response flow with a defined teacher-facing persona, a tool-use decision and a fallback when no suitable material is found.
Connect an API to a frontend flow
The role requires backend fundamentals in APIs and databases plus enough frontend ability to work end to end. Suggested exercise: expose retrieved material and an assistant response through a Python API, then build a minimal frontend interaction that displays results and handles loading and error states.
Instrument and document a cloud deployment
Solid AWS experience across compute, storage, databases, containers and monitoring is required, alongside clear technical documentation. Suggested exercise: diagram a plausible AWS deployment for the prototype, specify what you would monitor for failures and latency, and write a short operational handoff. This diagram is preparation, not a claim about Audrey’s architecture.
Use AI-assisted development critically
The posting explicitly mentions AI-assisted development tools such as Cursor. Suggested exercise: use one to draft a small pipeline or API change, then review the generated code, add tests and record corrections rather than accepting the output unverified.

Your preparation checklist

Prepare a production AI application walkthrough
The role requires hands-on experience shipping an LLM/AI application. If you have a relevant example, prepare a diagram and a concise account of your ownership, backend and frontend work, launch decisions, and measured outcomes.
Sketch Audrey’s retrieval pipeline
As a design exercise, map how videos, printables, guides, and multilingual content could be ingested, structured, embedded, searched, and re-ranked. Note how you would assess relevance and latency; these are proposed choices, not stated architecture.
Design a teacher-facing response flow
Prepare a sample request-to-response flow connecting a Python API, retrieval, LLM orchestration, and the product interface. Explain proposed tool use, persona, guardrails, and what happens when suitable material is not found.
Bring an evaluation and safety plan
Draft a small test set for lesson-planning and material-discovery requests, with proposed checks for recommendation quality, response speed, reliability, and privacy. Identify FERPA/COPPA questions to resolve rather than assuming a compliance design.
Summarize cloud and development practices
Prepare examples, where available, of AWS compute, storage, databases, containers, and monitoring decisions, plus how you review work produced with AI-assisted tools. Bedrock or self-hosted inference is a bonus, not a stated requirement.
Prepare a product and collaboration example
Choose an example, if available, showing how user needs changed an engineering decision. Bring a brief written update that demonstrates how you would share trade-offs and progress with product, engineering, and remote UK/US colleagues.
Write targeted questions for MarcoPolo Learning
Ask how educators currently find materials, what Audrey’s launch priorities and evaluation criteria are, and how AI ownership is divided across teams. Clarify the engagement terms: the description says full-time and permanent, while the page classifies the role as Contractor.

Availability last checked: 5 Nov 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 India. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…