New opportunity

Engineer III - App Engg & AI

Anblicks · Ahmedabad, India

About this role

Anblicks seeks an AI Developer for its Engineer III - App Engg & AI role in Ahmedabad. The position combines software engineering with large language models, retrieval, tool integration and agentic workflows to turn validated use cases into secure, observable, production-ready applications. The posting header lists “5 Years”; the required qualifications specify 3 to 5+ years of software engineering experience, including at least 1 year building AI/ML or LLM-integrated applications.

Responsibilities

Build LLM-enabled features, RAG pipelines, agentic workflows, intelligent automation and model/API integrations. Develop backend services, APIs, user-facing components, data integrations and cloud-native deployment assets as needed.
Implement prompt templates, grounding, structured outputs, tool calling, context management, model routing and evaluation harnesses. Connect securely to enterprise APIs, workflow systems, knowledge repositories, data platforms and approved external services.
Write unit, integration, regression, security and AI-quality tests covering groundedness, correctness, safety, latency and cost. Add logging, tracing, telemetry, exception handling, usage controls, runbooks and support diagnostics.
Follow secure coding, secrets management, least-privilege, privacy, data-handling and responsible AI requirements. Estimate and deliver sprint work, participate in design and code reviews, resolve defects, and document interfaces, deployments and operational guidance.

Required qualifications

Strong Python skills or equivalent proficiency in an enterprise AI application language; experience with REST APIs, asynchronous processing, authentication, data access, testing, Git and CI/CD.
Practical experience with LLM SDKs/APIs, prompt orchestration, RAG, vector retrieval, tool calling and structured outputs. Experience deploying on a major cloud platform and troubleshooting application, model, integration and data layers.
Ability to produce maintainable production code, participate in peer reviews and adapt to evaluation evidence and stakeholder feedback. Working knowledge of privacy, secure coding and responsible AI controls.

Preferred qualifications

Experience with Azure AI Foundry, Azure OpenAI, Anthropic Claude, Model Context Protocol or comparable platforms; full-stack frameworks, containers, serverless services, API management, event-driven integration and observability tools.
Familiarity with prompt/model evaluation frameworks, synthetic test generation, red-team testing and LLMOps. Experience with enterprise work-management, IT service-management, risk, compliance or knowledge platforms; regulated domains such as financial services, lending or auto finance; or SQL, data engineering, semantic layers, vector databases/search and knowledge graphs.

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

Skills for this role

PythonSoftware engineeringLarge language modelsRetrieval-augmented generationAgentic workflowsPrompt engineeringVector retrievalTool callingStructured outputsREST APIsAsynchronous processingAuthenticationGitCI/CDCloud deploymentAPI integrationUnit testingIntegration testingRegression testingSecurity testingLLM evaluationObservabilitySecure codingSecrets managementResponsible AIAzure AI FoundryAzure OpenAIAnthropic ClaudeModel Context ProtocolContainers Serverless services API management Event-driven integration SQL Data?

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

Turn validated use cases into production AI applications
Posting: Work with architects and product stakeholders to build LLM-enabled features, RAG pipelines, agentic workflows and model integrations using modular code. Preparation suggestion: Be ready to explain how a use case becomes a deployable application, including the choices you would validate before implementation.
Build secure, end-to-end integrations
Posting: Develop services, APIs, user-facing components and data integrations as needed; connect to enterprise systems while following authentication, secrets, least-privilege and privacy requirements. Preparation suggestion: Prepare an example of how you would define an integration boundary and protect data passed to an LLM or tool.
Evaluate AI behavior alongside software quality
Posting: Implement grounding, structured outputs, tool calling and evaluation harnesses; test correctness, groundedness, safety, latency and cost as well as conventional software behavior. Preparation suggestion: Describe how evaluation results would drive a change to retrieval, prompts or application code.
Make applications supportable and deliver iteratively
Posting: Add logging, tracing, telemetry, exception handling, usage controls and runbooks; deliver sprint work, review code, resolve defects and document operational guidance. Preparation suggestion: Explain how you would make an AI feature diagnosable after deployment without delaying essential testing or documentation.

Likely challenges to think through

Balancing answer quality with latency and cost
Preparation hypothesis, not a reported employer problem: More retrieval, context or model calls may improve groundedness while increasing response time and cost. Prepare to discuss how you would compare options using the posting’s quality, latency and cost evaluations.
Giving agents useful access without excessive permissions
Preparation hypothesis, not a reported employer problem: Tool calling across enterprise systems may require enough access to complete a workflow while limiting data exposure and unintended actions. Prepare to describe permission boundaries, input validation and tests for unsafe tool behavior.
Diagnosing failures across multiple layers
Preparation hypothesis, not a reported employer problem: An incorrect result could originate in application code, retrieval data, a model response or an external integration. Prepare a troubleshooting approach that uses traces and targeted tests to isolate the layer before changing prompts or code.

Practice your answers

Technical practice

How would you turn a validated AI use case into a production-ready application?

What to coverSuggested preparation: Describe how you would divide the work among LLM features, backend APIs, user-facing components, data integrations and cloud deployment assets as the solution requires. Explain how modular design, security, testing and observability influence those choices.

How would you design and evaluate a RAG pipeline for an enterprise knowledge repository?

What to coverSuggested preparation: Cover document access, retrieval, context selection and grounding the answer in retrieved material. Explain how you would investigate poor retrieval or unsupported answers.

How would you manage prompts, context and structured outputs across models?

What to coverSuggested preparation: Discuss prompt templates, context limits, output validation and when model routing might be useful. Explain how evaluation evidence would guide a change rather than relying on a single successful response.

How would you make an agentic workflow's tool calls reliable?

What to coverSuggested preparation: Describe tool selection, input validation, handling failed calls and limits on actions. Address how you would prevent an LLM from gaining broader access than the connected service permits.

How would you integrate an AI feature with enterprise APIs and data platforms?

What to coverSuggested preparation: Cover REST interfaces, authentication, data access and asynchronous processing where appropriate. Explain trade-offs around latency, retries and handling integration failures.

What would you measure before releasing an LLM-enabled feature?

What to coverSuggested preparation: Describe an evaluation set and checks for groundedness, correctness, safety, latency and cost. Explain how you would use failed cases to refine the implementation.

How would you test and deliver changes to an AI-integrated application?

What to coverSuggested preparation: Distinguish unit, integration, regression, security and AI-quality tests. Explain how Git and CI/CD could help catch changes to application behavior and model-dependent outputs.

What security and privacy controls would you apply to an AI application?

What to coverSuggested preparation: Discuss secrets management, least-privilege access, data handling and secure connectors. Explain how responsible-AI and security checks would apply to both application code and LLM interactions.

How would you troubleshoot a production answer that is incorrect or unexpectedly slow?

What to coverSuggested preparation: Trace the request across application, model, integration and data layers using logs and telemetry. Explain how you would isolate the cause while checking latency, usage and error patterns.

What would you include in a cloud deployment and support handoff?

What to coverSuggested preparation: Cover configuration, deployment procedures, exception handling, usage controls and support diagnostics. Explain what a runbook should tell someone responding to a failure.

Behavioural practice

When have you changed an implementation after evaluation or stakeholder feedback?

What to coverSuggested preparation: Give a real example using the situation, your action and the result. Explain what the evidence or feedback showed and how you decided what to change.

When have you managed a difficult commitment in a short iteration?

What to coverSuggested preparation: Describe the situation, how you estimated or reprioritized the work, and the result. Include how you kept delivery status transparent.

When has a design or code review improved your work?

What to coverSuggested preparation: Give a real example of the review situation, the feedback you acted on and the result. Focus on maintainability or production readiness rather than simply saying you accepted feedback.

When have you resolved a defect spanning more than one system layer?

What to coverSuggested preparation: Describe the symptoms, your investigation and the result. Make clear how you distinguished an application issue from a model, integration or data issue.

When have you documented a system so others could operate or extend it?

What to coverSuggested preparation: Give a real example using the situation, what you documented or transferred, and the result. Focus on interfaces, deployment steps, architecture decisions or operational guidance relevant to the role.

Sharpen your knowledge

Topics to prepare

RAG and grounding design
The posting requires RAG, vector retrieval and grounding. For revision, sketch how a question moves through retrieval, context selection and answer generation; explain how you would detect an answer unsupported by retrieved material.
Agent and tool-control boundaries
The role includes agentic workflows, tool calling, context management and model routing. For revision, diagram a workflow that calls an enterprise API and explain when it should request a tool call, reject one or stop.
AI-quality evaluation
The posting calls for evaluation harnesses and tests of groundedness, correctness, safety, latency and cost. For revision, define a small test set and explain what evidence would justify changing a prompt or implementation.
Enterprise security and responsible AI
Secure connectors, secrets management, least privilege, privacy and data handling are explicit responsibilities. For revision, review a hypothetical knowledge-repository integration and identify access-control, sensitive-data and unsafe-output failure points.
Production application design
The role spans backend services, APIs, asynchronous processing, cloud deployment and operational readiness. For revision, sketch the components of an LLM-enabled service and trace how a failed model or data call would be detected and handled.

Skills to revise

Implement an asynchronous API endpoint
Required qualifications include Python or an equivalent enterprise AI application language, REST APIs, asynchronous processing and authentication. As a revision exercise, implement a small authenticated endpoint that calls an LLM API asynchronously and handles timeouts and errors.
Build a minimal retrieval flow
Practical RAG and vector-retrieval experience is required. As a revision exercise, connect a small document collection to retrieval and generate answers from selected passages, checking whether each answer is supported.
Implement structured tool interactions
The posting explicitly names prompt templates, structured outputs and tool calling. As a revision exercise, create a prompt and tool-call flow for a sample API, validate the returned structure and handle invalid arguments without executing the call.
Write application and AI-quality tests
Unit, integration, regression, security and AI-quality tests are listed responsibilities. As a revision exercise, add tests for an API-to-model flow, including an authentication failure, a retrieval failure and an unsupported generated answer.
Prepare a deployable, diagnosable change
The posting requires cloud-deployment experience, Git and CI/CD, and calls for logging, tracing, telemetry and runbooks. As a revision exercise, package a small service for deployment on a major cloud platform, run its tests through CI/CD, and write a short diagnostic procedure for a failed request.

Your preparation checklist

Prepare an AI application walkthrough
The posting calls for LLM features, RAG, tool integration and production-ready code. Prepare one project walkthrough, if available, showing the use case, architecture, your contributions, trade-offs and results; otherwise use a clearly labeled design example.
Sketch a secure RAG and tool-calling design
As a preparation exercise, draw the flow from data ingestion and vector retrieval through grounding, structured outputs and tool calls. Explain authentication, least-privilege access, secrets handling and failure behavior.
Assemble engineering evidence
The role requires Python or equivalent enterprise AI development, REST APIs, asynchronous processing, testing, Git, CI/CD and cloud deployment. Identify specific examples you can explain, including a code or architecture decision and a cross-layer troubleshooting case.
Draft an evaluation plan
Prepare a small test matrix covering unit, integration, regression and security tests alongside groundedness, correctness, safety, latency and cost. Explain how evaluation evidence would guide a change to prompts, retrieval or code.
Walk through operational readiness
Prepare to explain what you would log and trace, how you would handle exceptions and usage controls, and what a runbook would tell support staff. These are explicit responsibilities in the posting.
Rehearse delivery and review examples
Select examples, if available, of estimating sprint work, responding to code-review feedback, resolving a defect and documenting an interface or deployment. Focus on decisions and changes made, not just the final result.
Bring focused questions for Anblicks
Ask which validated use cases the role will address first, what enterprise systems and cloud platform are involved, how AI quality is evaluated, and who owns deployment and support decisions.

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…