New opportunity

Senior - AI Engineer

iTech · Chennai, India

About this role

iTech seeks a hands-on Senior AI Engineer in Chennai to lead delivery of LLM and Document AI solutions, own production AI services, mentor engineers and work directly with clients in the US and India. This is a technical leadership role rather than a purely managerial position.

Responsibilities

Own AI features from requirements and solution design through development, deployment and monitoring. Build LLM and RAG applications using vector databases for document search, question answering, extraction and summarization.
Develop Document AI pipelines for OCR post-processing, entity and key-value extraction, classification, and layout and table understanding. Deploy scalable APIs and microservices in cloud and on-premises environments.
Lead the technical work on 2–3 AI projects, guide architecture and design, conduct code reviews, and coach junior engineers. Work with client stakeholders on requirements, solution presentations and reviews.
Account for security, privacy and cost, particularly in data-sensitive BFSI, healthcare, legal and HR applications.

Requirements

5–6 years of total experience, including at least 3 years in production AI, deep learning or NLP, and demonstrated delivery of AI solutions from proof of concept to production.
Strong Python and computer science fundamentals; hands-on experience with AI, NLP and LLM tasks such as classification, named entity recognition, question answering, summarization and document understanding. Experience with some of PyTorch, TensorFlow, Hugging Face, LangChain, LlamaIndex or spaCy.
Experience building RESTful ML services integrated with web or mobile applications; Docker, CI/CD and AWS or GCP; SQL and at least one relational database; and production MLOps practices including monitoring, logging, versioning, rollback, and performance and cost optimisation.
Excellent written and spoken English, client communication, ownership and mentoring skills. Willingness to overlap partially with US time zones for key meetings. A bachelor’s degree in Computer Science, Engineering or a related field is mandatory.

Preferred

Relevant industry experience; Vision Transformers or advanced computer vision for document processing; experiment tracking tools such as MLflow or DVC; security and compliance exposure; on-premises or VPC-isolated deployments; a systems language such as Go, Rust, C++ or Scala; and a master’s degree.

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

Skills for this role

PythonLarge Language ModelsRetrieval-Augmented GenerationDocument AIOCRNatural Language ProcessingDeep LearningNamed Entity RecognitionDocument ClassificationInformation ExtractionDocument Layout UnderstandingPyTorchTensorFlowHugging FaceLangChainLlamaIndexspaCypgvectorFAISSChromaOpenSearchREST APIsMicroservicesDockerCI/CDAWSGCPSQLPostgreSQLMySQL_coryn_POSTGRESQL_eeMySQL_IN_PSQL_PSQL_PSQL_PSQL_PSQL_PSQL_PSQL_PSQL_PSQL_P

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 AI delivery end to end
The role covers requirements, solution design, development, deployment and monitoring, with an explicit emphasis on taking AI work from proof of concept into production.
Build document-focused LLM and RAG solutions
Develop document search, question answering, extraction and summarization using LLMs, RAG and vector databases.
Deliver Document AI pipelines
Work on OCR post-processing, entity and key-value extraction, classification, and document layout and table understanding.
Run reliable, secure production services
Deploy scalable APIs and microservices in cloud and on-prem environments. Account for monitoring, logging, versioning, rollback, performance, cost, security and privacy.
Lead technical work and client discussions
Guide architecture and reviews across 2–3 AI projects, mentor junior engineers, and work directly with US and India client stakeholders on requirements and solution reviews.

Likely challenges to think through

Balancing extraction quality with document variation
Preparation hypothesis: OCR errors and varied layouts may make extraction quality difficult to maintain across document types. Prepare to explain how you would evaluate errors, choose between pipeline approaches and decide when a solution is ready for production.
Trading off deployment constraints and operating cost
Preparation hypothesis: cloud and on-prem deployments in data-sensitive domains may constrain design choices while services still need to scale. Prepare to discuss how you would compare privacy, security, latency and cost, then monitor and roll back a deployed service.
Staying hands-on across multiple projects
Preparation hypothesis: technical leadership on 2–3 projects alongside client reviews and mentoring may compete with implementation time. Prepare an example of how you would prioritize architectural decisions, delegate work and communicate trade-offs without losing delivery ownership.

Practice your answers

Technical practice

How would you design a document-search and question-answering RAG service?

What to coverSuggested preparation: Explain document ingestion, chunking, embeddings, retrieval, answer generation and citations. Discuss how you would choose among the posting’s example vector stores based on scale, filtering and operational needs.

How would you evaluate a RAG application before and after deployment?

What to coverSuggested preparation: Distinguish retrieval quality from answer accuracy and faithfulness using representative document questions. Explain how you would investigate failures such as missed passages, unsupported answers and poor summarization.

How would you build a pipeline for extracting information from scanned, structured and layout-heavy documents?

What to coverSuggested preparation: Cover OCR post-processing, entity and key-value extraction, classification, and layout or table understanding. Discuss error propagation and when layout-aware methods would be worth their added complexity.

How would you choose an approach for document classification versus named entity recognition?

What to coverSuggested preparation: Compare the labels, training data and evaluation measures each task needs. Explain the trade-offs between simpler baselines and deep-learning approaches without assuming one model fits every document type.

How would you expose an AI pipeline through a REST API for a web or mobile application?

What to coverSuggested preparation: Address request and response design, input validation, latency, failure handling and scaling. Explain when a long-running document task should be handled asynchronously rather than in a single request.

How would you take an AI service from a proof of concept to cloud or on-prem production?

What to coverSuggested preparation: Describe reproducible packaging with Docker, CI/CD checks, deployment configuration and operational readiness. Compare cloud and on-prem constraints without assuming a particular client infrastructure.

How would you monitor and safely update a production AI service?

What to coverSuggested preparation: Discuss service logs and performance alongside model-quality signals, versioning and rollback. Explain how you would detect whether a problem comes from the API, document pipeline, retrieval or model.

How would you balance privacy, security and cost in a document AI design?

What to coverSuggested preparation: Consider sensitive inputs, access controls, data retention and where processing occurs in domains named by the posting. Explain trade-offs among isolation, operational complexity, latency and usage cost.

Where would SQL and a relational database fit in a document AI service?

What to coverSuggested preparation: Discuss storing document metadata, processing status and application records alongside a retrieval system. Explain how you would keep records consistent when processing or indexing fails.

How would you structure Python code for a maintainable AI pipeline?

What to coverSuggested preparation: Cover clear interfaces between ingestion, inference and serving, plus testing and clean-code practices. Discuss data-structure or algorithm choices where they affect throughput or memory use.

Behavioural practice

Tell me about a time you clarified ambiguous requirements with a client.

What to coverSuggested preparation: Use a real situation, the questions and trade-offs you raised, and the resulting agreement or change in scope. Focus on how you translated a business need into a technical plan.

Tell me about a time you guided another engineer through a design or code-review issue.

What to coverSuggested preparation: Describe the situation, your coaching actions and the result. Show how you maintained technical standards while helping the engineer make progress.

Tell me about a time you balanced competing priorities across technical projects.

What to coverSuggested preparation: Describe the situation, how you assessed risks and dependencies, and what you communicated before acting. Include the result and any trade-off you would handle differently now.

Tell me about a time you took ownership of a solution beyond its prototype.

What to coverSuggested preparation: Use a real example to explain the production challenge, your deployment or monitoring actions, and the result. Distinguish your contribution from the work of the wider team.

Tell me about a time you explained a complex AI decision to non-technical stakeholders.

What to coverSuggested preparation: Describe the audience and decision, how you made limitations and trade-offs understandable, and the outcome. An example involving a client review or distributed team is relevant, but not required.

Sharpen your knowledge

Topics to prepare

RAG architecture and retrieval quality
Suggested revision exercise: Sketch a document search and question-answering system, then explain how you would assess retrieval quality and answer accuracy. The posting explicitly calls for LLM/RAG solutions using vector databases for search, Q&A, extraction and summarization.
Document AI pipeline design
Suggested revision exercise: Trace an invoice or legal document through OCR post-processing, entity and key-value extraction, classification, and layout or table understanding; identify likely error points. These pipeline stages are explicit responsibilities.
NLP task selection and evaluation
Suggested revision exercise: For classification, named entity recognition and summarization, explain what outputs and failure cases you would review before release. The posting requires hands-on AI/NLP work across these tasks.
Production lifecycle and MLOps
Suggested revision exercise: Diagram a proof-of-concept-to-production release, including logging, monitoring, model versioning, rollback, and performance and cost checks. The role owns production AI services and explicitly requires these MLOps practices.
Security, privacy and deployment trade-offs
Suggested revision exercise: Compare cloud and on-prem deployment options for a data-sensitive document workflow, noting privacy and cost questions to resolve. Both environments are in scope, and the posting asks engineers to design with security, privacy and cost in mind; a particular compliance framework is not required.

Skills to revise

Python implementation and code review
Suggested revision exercise: Implement a small document-classification or entity-extraction component in Python, then review it for data structures, algorithmic choices and clean code. Strong Python and core computer science fundamentals are explicit requirements.
Build a document RAG workflow
Suggested revision exercise: Create a small search-and-answer workflow over sample documents using one listed vector database, such as FAISS or pgvector, and explain how you would investigate poor answers. The posting explicitly names document RAG and vector databases.
Implement a Document AI pipeline
Suggested revision exercise: Process sample OCR output into structured key-value fields and flag cases where layout or tables change the interpretation. OCR post-processing, extraction and layout/table understanding are stated responsibilities.
Package and release an ML API
Suggested revision exercise: Expose one inference task through a RESTful Python service, containerize it with Docker, and outline CI/CD and deployment considerations for AWS or GCP. The posting requires deployed RESTful ML services, Docker, CI/CD and experience with either cloud platform.
Query and validate relational data
Suggested revision exercise: Write SQL queries to retrieve and check document-processing results in a relational database, and explain how the service would use them. Good SQL skills and experience with at least one RDBMS are required.
Lead a client-facing design review
Suggested revision exercise: Present a proposed AI feature in plain English, clarify requirements, defend an architecture choice and give constructive code-review feedback. The role explicitly involves client reviews, technical leadership across 2–3 projects and mentoring junior engineers.

Your preparation checklist

Prepare an end-to-end delivery walkthrough
Select a real AI project you can discuss and outline requirements, proof of concept, production deployment, monitoring, results and your own decisions. The role calls for hands-on ownership beyond notebooks.
Sketch a document AI pipeline
Prepare a design exercise covering OCR post-processing, entity or key-value extraction, classification, and layout or table handling. State how you would measure quality and handle errors; these are preparation choices, not a specified interview task.
Design a document-search and RAG service
Draw a proposed flow from document ingestion and retrieval through question answering or summarization. Explain vector-database choices and how you would evaluate answers, privacy and cost.
Assemble production engineering evidence
Prepare concrete examples of Python RESTful ML APIs, web or mobile integration, Docker, CI/CD, AWS or GCP, SQL and a relational database. Be ready to explain logging, versioning, rollback, performance and cost decisions.
Prepare a technical-leadership example
Outline how you guided architecture, reviewed code or mentored engineers across AI work. Separate your hands-on contributions from team outcomes; the advertised role leads technical work on 2–3 projects.
Rehearse a client-facing requirements discussion
Practice explaining an AI design and its trade-offs in plain English, then prepare clarifying questions for a data-sensitive client. The role involves US and India stakeholders and partial US time-zone overlap for key meetings.
Write questions for the employer
Ask how the 2–3 projects are staffed, which document and LLM use cases are priorities, what production monitoring looks like, and when cloud versus on-prem deployment is needed.

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…