New opportunity

Senior ML Engineer

Bot Consulting Private Limited · Jaipur, India

About this role

Bot Consulting Private Limited seeks a Senior ML Engineer for client-facing AI and Generative AI delivery on AWS for AllCloud. This hands-on consulting role owns technical workstreams from discovery and proof-of-concept through production hardening, including AWS Migration Acceleration Program engagements. The engineer works with client stakeholders, architects, data scientists and engineers across the US and India.

Responsibilities

Design and deploy Amazon Bedrock applications for retrieval-augmented generation, document intelligence, summarisation, classification and assistants. Build tool-using and multi-agent systems, and retrieval pipelines with embeddings, hybrid search, relevance tuning, reranking, citations and grounding.
Own SageMaker workflows spanning feature engineering, training, tuning, model registry, deployment, retraining, drift detection and CI/CD. Establish evaluation and observability using golden datasets, LLM-as-judge, regression tests, tracing and human review.
Implement guardrails, prompt-injection defences and PII controls. Integrate AWS data services; design secure, multi-tenant, scalable and cost-aware architectures; deliver infrastructure as code; and optimise model, token, compute and storage costs.
Lead client discovery, requirements and architecture discussions, presentations and executive readouts. Contribute code reviews, reusable engineering patterns and mentoring.

Required qualifications

Bachelor’s or Master’s degree in a quantitative field, or equivalent demonstrated experience; 5–8+ years in ML engineering, data science or AI engineering, including at least 3 years of hands-on AWS experience and at least 1 year of production-grade GenAI/LLM experience. Candidates must be able to explain a production system or client-accepted proof-of-concept in architectural detail.
Hands-on Bedrock, RAG, agentic systems, OpenSearch, SageMaker and MLOps experience; AWS data, architecture and security knowledge; expert Python and strong SQL, Git, Docker and Linux skills. A foundation in statistics, classical ML, forecasting and NLP is required.
Experience leading sessions with external clients or business stakeholders, working across significant time-zone differences, and producing clear English documentation and senior-audience presentations. AWS ML certification is required at hire or within 60 days.

Preferred qualifications include Anthropic certification, AWS Solutions Architect – Professional certification, consulting or AWS Partner experience, relevant regulated-industry experience, agent and retrieval frameworks, knowledge graphs, additional data platforms, and technical publications or community contributions. The posting mentions a competitive salary, wellness packages, professional development, mentorship and flexible work arrangements, without specifying a salary or work mode.

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

Skills for this role

Generative AIAmazon BedrockRetrieval-Augmented GenerationAmazon Bedrock Knowledge BasesAmazon Bedrock GuardrailsFoundation modelsAgentic AIMulti-agent systemsAmazon Bedrock AgentsAmazon Bedrock AgentCoreStrands Agents SDKModel Context ProtocolAWS Step FunctionsAWS LambdaAmazon OpenSearchVector searchHybrid searchEmbeddingsBM25RerankingpgvectorDocument parsingPrompt engineeringAmazon SageMakerMLOpsModel monitoringDrift detectionLoRALLM evaluationLLM observability

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

Own client-facing delivery
Lead discovery, requirements and architecture discussions, then carry technical workstreams from proof-of-concept through production hardening. The role also calls for executive readouts, clear architecture documentation and collaboration across the US and India.
Build Bedrock applications and retrieval systems
Design and deploy RAG, document intelligence and assistant solutions, including tool-using and multi-agent systems. Retrieval work spans parsing, chunking, hybrid search, reranking, citations and grounding.
Make AI behavior measurable and controlled
Establish evaluation and observability with golden datasets, regression tests, tracing, error taxonomies and human review. Implement guardrails, prompt-injection defences and PII handling for data-sensitive use cases.
Own production ML lifecycles
Run SageMaker workflows across feature engineering, training, tuning, model registry, deployment, retraining, drift detection and CI/CD—not just initial model development.
Engineer secure, cost-aware AWS delivery
Integrate AWS data services; design scalable, multi-tenant architectures; deliver infrastructure as code; and estimate and optimise model, token, compute and storage costs.

Likely challenges to think through

Balancing retrieval quality against operating cost
A likely design trade-off is improving grounded answers through parsing, hybrid search and reranking without unacceptable latency or cost. Prepare to explain how you would evaluate retrieval and answer quality, then choose model and pipeline components against those constraints.
Turning an accepted proof-of-concept into production
The posting spans discovery, POCs and production hardening, so a likely difficulty is preserving a useful prototype while adding deployment controls, monitoring and repeatable evaluation. Prepare a concrete architectural walkthrough that distinguishes POC decisions from production changes.
Reconciling client needs with data controls
Client-facing, multi-tenant and data-sensitive work can create tension between useful AI outputs and isolation, PII protection and prompt-injection defences. Prepare to discuss where you would place controls, how you would test them and which decisions require stakeholder agreement.

Practice your answers

Technical practice

How would you select a Bedrock model and architecture for a client-facing document assistant?

What to coverThe role includes Bedrock assistants and model selection based on quality, latency, cost, and data residency. Suggested preparation: compare options against the client's requirements and explain what you would validate in a proof-of-concept.

How would you diagnose poor answer relevance in a RAG application?

What to coverThe posting names parsing, chunking, embeddings, hybrid search, reranking, citations, and grounding. Suggested preparation: trace failures from source documents through retrieval to the final answer, and explain how you would evaluate each stage.

When would you use a tool-using or multi-agent design rather than a simpler workflow?

What to coverThe role involves Bedrock agentic systems, Step Functions, and Lambda. Suggested preparation: discuss tool permissions, orchestration, failure handling, and whether the added complexity improves the use case.

How would you evaluate and monitor a production GenAI application?

What to coverThe posting specifies golden datasets, LLM-as-judge, regression tests, error taxonomies, tracing, and human review. Suggested preparation: explain how these methods complement one another and how findings would drive releases or fixes.

How would you defend a RAG assistant against prompt injection and sensitive-data exposure?

What to coverThe role calls for guardrails, prompt-injection defences, PII handling, and responsible AI controls. Suggested preparation: address untrusted retrieved content, access boundaries, testing, and the limits of any single control.

How would you manage a model through the SageMaker lifecycle?

What to coverThe posting covers feature engineering, training, tuning, Model Registry, Pipelines, deployment, monitoring, retraining, and drift detection. Suggested preparation: describe versioning, promotion criteria, CI/CD checks, and how you would distinguish drift from an isolated failure.

How would you build a reliable data foundation for an AI workload on AWS?

What to coverThe role involves integrating services including S3, Glue, Spark, Athena, Redshift, DMS, Kinesis, and Firehose. Suggested preparation: choose a plausible batch or streaming scenario and explain data quality, freshness, and service trade-offs without assuming a specific client stack.

How would you design and deliver a secure, multi-tenant AI service?

What to coverThe posting names IAM, Lake Formation, VPC, PrivateLink, API Gateway, compute services, infrastructure as code, and CI/CD. Suggested preparation: explain tenant isolation, network and data access, deployment controls, and how you would test the architecture.

How would you estimate and reduce the cost of an AI application?

What to coverThe role explicitly covers model selection, token economics, caching, compute, and storage costs. Suggested preparation: show how you would measure cost alongside quality and latency, then justify changes without assuming they preserve output quality.

How would you establish a baseline for a forecasting or NLP task before using a more complex model?

What to coverThe requirements include statistics, classical ML, time-series forecasting, and NLP tasks such as classification and entity extraction. Suggested preparation: discuss baselines, appropriate validation, error analysis, and when added model complexity would be warranted.

Behavioural practice

Tell me about a time you turned ambiguous client needs into a technical plan.

What to coverThis role leads discovery, requirements sessions, and architecture discussions. Suggested preparation: use a real situation, your questions and decisions, and the resulting agreed scope or next step.

Describe a time you explained a technical trade-off to senior stakeholders.

What to coverThe posting calls for client presentations, executive readouts, and clear architecture documentation. Suggested preparation: give the situation, how you adapted your explanation and recommendation, and the decision or feedback that followed.

Give an example of coordinating delivery across significant time-zone differences.

What to coverThe role works with teams and stakeholders across the US and India and requires self-direction. Suggested preparation: describe the coordination problem, your communication and handoff actions, and their effect on delivery.

Tell me about a time you improved another engineer's work through review or mentoring.

What to coverCode reviews, reusable engineering patterns, and mentoring are stated responsibilities. Suggested preparation: explain the situation, the specific guidance or pattern you contributed, and the observable result.

Describe a time you moved a proof-of-concept toward a dependable production system.

What to coverThe role spans proof-of-concepts through production hardening and owns technical workstreams end to end. Suggested preparation: use a real example to cover the initial constraints, your prioritisation and validation steps, and the resulting outcome.

Sharpen your knowledge

Topics to prepare

Bedrock application and model selection
The role requires Bedrock applications for document intelligence, summarisation, classification and assistants. Suggested exercise: compare two foundation-model choices for one use case, explaining quality, latency, cost and data-residency trade-offs.
Retrieval and grounding
The posting specifies document parsing, chunking, embeddings, BM25, hybrid search, reranking, citations and retrieval evaluation. Suggested exercise: trace a document-to-answer RAG design and explain how you would diagnose an irrelevant or unsupported answer.
Agentic system design
Tool-using and multi-agent systems are required, with Bedrock, Step Functions, Lambda and Bedrock Agents/AgentCore named. Suggested exercise: diagram a tool-calling workflow, including where orchestration, failure handling and human review belong.
SageMaker, MLOps and ML foundations
The role covers training through deployment, retraining and drift detection, alongside statistics, classical ML, forecasting and NLP. Suggested exercise: explain how you would select a baseline, evaluate a forecasting or classification model, and decide when drift warrants retraining.
Evaluation and responsible AI
Golden datasets, LLM-as-judge, regression tests, tracing, guardrails, prompt-injection defences and PII handling are explicit. Suggested exercise: design an evaluation plan that separates retrieval failures, answer-quality failures and safety failures.
AWS architecture, security and cost
The posting calls for secure, multi-tenant, scalable AWS designs, data-service integration and cost optimisation. Suggested exercise: review an AI architecture for IAM and network boundaries, data access, token and compute costs, and scaling trade-offs. The AWS ML certification is required at hire or within 60 days.

Skills to revise

Implement a grounded RAG flow
Suggested exercise: use Python and the posting's Bedrock and OpenSearch stack to sketch or build document ingestion, hybrid retrieval, reranking and cited answers; test retrieval against a small labelled query set. This rehearses the specified hands-on retrieval work.
Build and inspect a tool-using workflow
Suggested exercise: implement a small Bedrock tool-calling flow with Lambda or Step Functions, then test tool errors and record the execution trace. This reflects the required agentic delivery work.
Practice a SageMaker delivery path
Suggested exercise: walk through or implement feature preparation, training, tuning, Model Registry, deployment and a monitoring trigger, showing where CI/CD and retraining fit. These are named lifecycle responsibilities.
Test quality and safeguards
Suggested exercise: create a small golden dataset and regression tests, then probe a Bedrock application for unsupported claims, prompt injection and PII exposure; document human-review criteria. This targets the posting's evaluation and responsible-AI requirements.
Review infrastructure and workload costs
Suggested exercise: draft an AWS CDK, CloudFormation or Terraform outline for a secure, multi-tenant AI service and estimate model-token, compute and storage cost drivers. Infrastructure as code and cost optimisation are explicit responsibilities.
Present a technical workstream
Suggested exercise: prepare a concise architecture diagram and senior-audience readout for a production system or client-accepted proof-of-concept, covering requirements, decisions, evaluation and trade-offs. The role requires client-led sessions and an architectural explanation of such a system.

Your preparation checklist

Prepare a production or accepted-POC walkthrough
Create a diagram and brief narrative for a GenAI system you can discuss: client need, architecture, your decisions, evaluation, deployment and trade-offs. The posting explicitly asks for architectural detail on a production system or client-accepted proof of concept.
Practice a Bedrock design exercise
Sketch a document-based assistant using Bedrock, parsing, chunking, hybrid OpenSearch retrieval, reranking and grounded citations. Explain when you would add tools or multiple agents, and how you would choose a model for quality, latency, cost and data residency.
Map an end-to-end SageMaker workflow
Prepare a lifecycle diagram covering feature engineering, training and tuning, Model Registry, deployment, CI/CD, monitoring, drift detection and retraining. Be ready to explain where classical ML or forecasting is more suitable than an LLM.
Build an evaluation and safety plan
Draft a small example golden dataset and a test plan for retrieval quality, answer regressions, tracing and human review. Include prompt-injection defences, PII handling and guardrails, reflecting the posting’s data-sensitive use cases.
Rehearse a secure, cost-aware AWS architecture
Draw a proposed multi-tenant deployment with data access, IAM and network boundaries, infrastructure as code, monitoring and CI/CD. Annotate how you would estimate and control token, model, compute and storage costs; treat the diagram as a practice design, not a claimed company architecture.
Gather hands-on engineering evidence
Select code or examples you can discuss for Python, SQL, testing, Git, Docker and AWS integration. Prepare concise explanations of design choices, code-review trade-offs and reusable patterns, all called out in the role.
Prepare a client-facing discovery exercise
Draft questions to clarify business goals, data constraints, acceptance criteria and the boundary between a proof of concept and production hardening. Practice a short architecture readout for a senior audience and how you would coordinate decisions across US and India time zones.
Check certification plans and ask focused questions
Review your plan for the required AWS ML certification, which the posting allows at hire or within 60 days; it lists Anthropic certification as preferred. Ask the employer how work is split among discovery, MAP engagements and production delivery, and how technical ownership and success are defined.

Availability last checked: 4 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…