New opportunity

Mactores - Generative AI Engineer

Mactores · Mumbai, India

About this role

Mactores is hiring a Generative AI Engineer in Mumbai to build production-grade applications using large language models (LLMs). The posting lists 3–7 years of experience and identifies India as the applicant location.

Responsibilities

Design and implement generative AI solutions, apply prompt engineering, and build scalable retrieval-augmented generation (RAG) systems.
Fine-tune and optimize models for performance, cost, and reliability; use AWS Bedrock, SageMaker, and Lambda for deployment and inference.
Develop APIs and backend components. Implement observability, performance monitoring, and security practices.
Support responsible AI adoption through evaluation, bias detection, and compliance.

Requirements

3+ years of Python experience and strong software engineering fundamentals. The structured posting also lists a bachelor's degree.
Hands-on LLM and prompt engineering experience; experience designing RAG pipelines and working with vector databases.
Proficiency in model fine-tuning, such as LoRA, and embedding-based systems; experience with cloud platforms and production AI deployments.
Strong debugging, optimization, problem-solving, and technical communication skills, with attention to cost, reliability, and performance.

Preferred

Practical experience with LangChain or LlamaIndex; exposure to multimodal AI; familiarity with ML/MLOps and large-scale deployment; and experience supporting high request volumes.

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

Skills for this role

Generative AILarge Language ModelsPrompt EngineeringRetrieval-Augmented GenerationPythonVector DatabasesModel Fine-TuningLoRAEmbeddingsAWSAmazon BedrockAmazon SageMakerAWS LambdaAPI DevelopmentBackend DevelopmentObservabilityPerformance MonitoringAI EvaluationBias DetectionResponsible AILangChainLlamaIndexMultimodal AIMLOpsDebuggingPerformance OptimizationTechnical Communication

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

Build scalable RAG systems
The posting calls for prompt engineering, RAG pipelines, vector databases, and embedding-based systems. Prepare to explain design choices for retrieval, grounding, and response quality.
Deliver production AI applications
The role includes Python-based APIs and backend components, plus deployment and inference using AWS Bedrock, SageMaker, and Lambda. Prepare to discuss how an LLM application moves from implementation to a reliable service.
Optimize model performance, cost, and reliability
Fine-tuning, including approaches such as LoRA, and model optimization are explicit requirements. Prepare to explain how you would measure improvements and choose between prompt, retrieval, and model changes.
Operate and safeguard deployed systems
The posting specifies observability, performance monitoring, security, evaluation, bias detection, and compliance. Prepare to describe what you would monitor and how evaluation would inform changes after deployment.

Likely challenges to think through

Balancing answer quality against serving cost
Preparation hypothesis: Better retrieval or model outputs may increase latency and cost, while cheaper settings may reduce quality. Prepare a comparison framework using evaluation results alongside latency, cost, and reliability measures.
Keeping RAG dependable as it scales
Preparation hypothesis: At higher request volumes, retrieval quality and response time may be difficult to maintain together. Prepare to discuss how you would diagnose failures across embeddings, vector search, prompts, and backend components; high-volume experience is listed as preferred, not required.
Turning responsible-AI goals into operational checks
Preparation hypothesis: Evaluation, bias detection, and compliance can be hard to reconcile with fast iteration. Prepare to explain how you would test changes, monitor outcomes, and decide when a release needs further review without assuming a specific Mactores policy.

Practice your answers

Technical practice

How would you design and evaluate prompts for a production LLM application?

What to coverThe posting calls for hands-on prompt engineering. Suggested preparation: explain how you would test prompts against representative inputs, check output quality and consistency, and weigh quality against latency and cost.

How would you design a scalable retrieval-augmented generation system?

What to coverThe role includes building scalable RAG systems. Suggested preparation: walk through ingestion, retrieval, context assembly and generation, then explain how you would evaluate answer quality and identify bottlenecks.

How would you choose and troubleshoot embeddings and vector retrieval?

What to coverThe posting specifies embedding-based systems and vector databases. Suggested preparation: discuss retrieval relevance, indexing and query choices, and how you would diagnose cases where useful source material is not retrieved.

When would you fine-tune an LLM with LoRA rather than improve its prompts or retrieval?

What to coverModel fine-tuning, including LoRA, is listed in the requirements. Suggested preparation: compare the approaches by the problem they address, the evidence needed to justify fine-tuning, and their performance, cost and maintenance implications.

How would you choose among Bedrock, SageMaker and Lambda for deployment and inference components?

What to coverThe posting names these AWS services for deployment and inference. Suggested preparation: describe a plausible division of responsibilities and explain how workload needs, latency, operational effort and cost would affect your choices; do not assume a prescribed architecture.

How would you structure a Python API for a production AI application?

What to coverThe role calls for Python fundamentals, APIs and backend components. Suggested preparation: cover request validation, error handling, testing and the boundary between the API, retrieval and model inference.

How would you diagnose a production LLM application that has become slow or unreliable?

What to coverThe posting emphasizes debugging, observability and performance monitoring. Suggested preparation: explain what you would measure across the API, retrieval and inference stages, and how those observations would guide a fix.

How would you manage cost, latency and reliability as request volume grows?

What to coverThe posting emphasizes production optimization and prefers experience with high request volumes. Suggested preparation: compare capacity, concurrency and inference-cost trade-offs, and explain how you would verify that an optimization has not reduced answer quality.

What security concerns would you address in an LLM-backed API and RAG system?

What to coverSecurity practices are part of the advertised responsibilities. Suggested preparation: discuss access to retrieved content, protection of sensitive inputs and outputs, and testing for attempts to manipulate instructions through untrusted content.

How would you evaluate a generative AI solution for quality, bias and compliance concerns?

What to coverThe posting explicitly includes evaluation, bias detection and compliance in responsible AI adoption. Suggested preparation: describe representative test cases, review of uneven or harmful outcomes, and how findings would influence release decisions or mitigations.

Behavioural practice

Tell me about a time you debugged a difficult production issue.

What to coverThe posting values debugging and production reliability. Suggested preparation: use a real example to explain the situation, your investigation and actions, and the observed result.

Describe a time you had to balance performance, reliability and cost.

What to coverThese trade-offs are explicit priorities in the role. Suggested preparation: explain the situation, the options and evidence you considered, the choice you made, and its result.

Tell me about a time you explained a technical AI or software decision to others.

What to coverThe posting calls for clear technical communication. Suggested preparation: identify the audience and situation, show how you explained the trade-offs or uncertainty, and describe the resulting decision or understanding.

Describe a time you found a quality, bias or security concern in an AI system.

What to coverThe role includes evaluation, bias detection and security practices. Suggested preparation: use a real situation to explain how you identified the concern, what action you took, and the outcome without overstating what was resolved.

Tell me about a time you moved an AI or backend component toward production use.

What to coverThe posting emphasizes production-grade applications and deployment. Suggested preparation: describe the starting state, your implementation or release actions, and the evidence you used to assess the result.

Sharpen your knowledge

Topics to prepare

LLM prompting strategies
Posting: Hands-on LLM and prompt engineering experience is required. Revision exercise: Compare two prompt designs for the same task and explain how you would judge output quality and consistency.
RAG, embeddings, and vector retrieval
Posting: The role calls for scalable RAG systems, vector databases, and embedding-based systems. Revision exercise: Diagram the path from document ingestion to retrieved context and answer generation; identify where retrieval quality or scale could suffer.
Fine-tuning trade-offs
Posting: Model fine-tuning, including LoRA as an example, and optimization for performance, cost, and reliability are stated. Revision exercise: Explain when you would test fine-tuning rather than prompt changes or RAG, and what you would measure before choosing.
AWS inference and deployment design
Posting: Bedrock, SageMaker, and Lambda are named for deployment and inference. Revision exercise: Sketch a production AI request flow using the relevant services and explain the deployment and cost trade-offs; the posting does not prescribe a specific architecture.
Production and responsible-AI safeguards
Posting: Observability, performance monitoring, security, evaluation, bias detection, and compliance are responsibilities. Revision exercise: Define what you would monitor and evaluate for a deployed LLM feature, including a way to investigate unreliable or biased outputs.

Skills to revise

Build a Python AI API
Posting: Python software engineering, APIs, and backend components are required. Suggested exercise: Implement a small Python endpoint that accepts a request, invokes an LLM component, handles errors, and returns a clear response; explain how you would debug failures.
Prototype and test a RAG pipeline
Posting: Designing RAG pipelines and working with vector databases are explicit requirements. Suggested exercise: Index a small document set with embeddings, retrieve relevant passages, and test questions whose answers are present and absent; inspect retrieval errors.
Run a prompt evaluation
Posting: Prompt engineering and AI evaluation are stated. Suggested exercise: Create a small set of representative inputs, run two prompt variants, and record output quality, failure cases, and inference cost.
Compare a LoRA fine-tuning experiment
Posting: Fine-tuning proficiency, with LoRA given as an example, is requested. Suggested exercise: Outline or run a small LoRA experiment against an unchanged model, then compare task quality, inference performance, and cost without assuming improvement.
Plan deployment and operational checks
Posting: AWS deployment, observability, monitoring, security, and reliability are stated. Suggested exercise: For a Bedrock or SageMaker inference path involving Lambda where appropriate, specify request metrics, error logging, access controls, and a test for degraded performance. This is a preparation scenario, not a stated Mactores design.

Your preparation checklist

Prepare a production AI project walkthrough
Choose a project you can discuss accurately. Map your Python implementation, LLM and prompt choices, API or backend design, deployment, and your own contribution to the responsibilities in Mactores’ posting.
Sketch a scalable RAG design
As a preparation exercise, draw the flow from documents through embeddings and a vector database to retrieval and answer generation. Explain how you would test retrieval quality, handle failures, and manage latency and cost.
Compare model-improvement options
Prepare a short example or decision table showing when you would refine prompts, change retrieval, or fine-tune with an approach such as LoRA. State how you would measure performance, cost, and reliability rather than assuming fine-tuning is always needed.
Practice an AWS deployment design
Outline how you would connect an AI inference service to a Python API, using Bedrock, SageMaker, and Lambda where appropriate. Be ready to explain your choices and trade-offs; the posting names these services but does not prescribe one architecture.
Assemble operational and responsible-AI evidence
Prepare examples you can substantiate of debugging, monitoring, security, evaluation, or bias detection. For a design exercise, specify a few metrics, failure checks, and compliance questions you would raise before production use.
Review preferred experience separately
If relevant to your experience, prepare brief examples involving LangChain or LlamaIndex, multimodal AI, MLOps, or high request volumes. Identify these as preferred qualifications, not core requirements you must claim to have met.
Write questions for the employer
Bring questions about the expected RAG workloads, evaluation criteria, AWS deployment choices, request volumes, and ownership of monitoring and responsible-AI reviews. These clarify work described in the posting without assuming Mactores’ internal processes.

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…