New opportunity

Solutions Architect - Applied AI

Lyzr AI · Bangalore, India

About this role

Lyzr AI seeks a Solutions Architect - Applied AI to design and deliver enterprise AI agents and workflows. The role combines AI architecture, model and data strategy, hands-on implementation, and consultation with internal and client teams. The stated location is Bangalore/Hybrid or Remote, and the position is full-time.

Responsibilities

Consult with teams to identify use cases and define agent capabilities, intents, interaction flows, and multi-agent orchestration architecture.
Guide selection, fine-tuning, and integration of LLMs, RAG pipelines, and transformer-based models. Define how models handle text, audio, video, and structured data; benchmark performance and assess retraining needs.
Advise on structured and unstructured data preparation and enrichment, including entity extraction, topic modeling, and summarization. Design data pipelines that support continuous improvement.
Lead prototypes and MVPs through to production-grade solutions. Work with developers on scalable AI services and interfaces, and write modular code to integrate AI components.
Explain technical choices to non-technical stakeholders and advise on product capabilities, limitations, and direction while owning delivery outcomes.

Required experience and skills:

At least 8 years of overall experience. The posting specifies 3–4 years in ML/advanced NLP and 1–2 years in LLMs; its detailed requirements also specify 3–4 years of core AI/ML expertise, 2–3 years of NLP and data intelligence, and 1–2 years in the LLM and agent ecosystem.
Machine learning and deep learning experience, including training and evaluating models and developing enterprise ML applications. Advanced NLP knowledge and structured and unstructured data feature engineering are required.
Understanding of LLM architecture, prompt engineering, and RAG frameworks, plus exposure to multimodal agent delivery. Strong problem-solving, communication, stakeholder management, and end-to-end ownership are sought.

Preferred qualifications

Experience with cloud ML platforms on AWS, GCP, or Azure; familiarity with agent orchestration platforms or LangChain-like frameworks; and understanding of vector databases, embeddings, and context-aware retrieval.

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

Skills for this role

AI agent architectureMulti-agent orchestrationMachine learningDeep learningLarge language modelsNatural language processingRetrieval-augmented generationTransformer modelsModel fine-tuningModel evaluationData preprocessingEntity extractionTopic modelingSummarizationFeature engineeringPrompt engineeringMultimodal AIData pipelinesPrototypingSoftware developmentStakeholder managementAWSGoogle Cloud PlatformMicrosoft AzureLangChainVector databasesEmbeddingsContext-aware retrieval

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

Define the right agent workflows
The role consults with internal and client teams to identify use cases, define agent capabilities and interaction flows, and design multi-agent orchestration. Prepare to explain how business goals become clear workflow boundaries and architecture decisions.
Make evidence-based model choices
The posting calls for selecting and integrating LLMs, RAG pipelines, and transformer models, then benchmarking performance and assessing fine-tuning or retraining needs. Prepare to explain what evidence would justify each approach.
Build a data strategy for improvement
The architect advises on structured and unstructured data preparation, oversees NLP enrichment, and designs pipelines for continuous improvement. Prepare to connect data quality and enrichment choices to model performance.
Own the path from MVP to production
The role leads prototypes into production-grade solutions, collaborates on scalable services and interfaces, and writes modular integration code. Prepare to show how architectural decisions translate into implementation and delivery.
Communicate limitations and recommendations
The posting expects actionable explanations for non-technical stakeholders, strategic input on product capabilities and limitations, and end-to-end accountability. Prepare to explain technical choices in terms of business goals and constraints.

Likely challenges to think through

Choosing orchestration only when it helps
Likely preparation challenge, not a stated employer problem: multi-agent designs can add coordination complexity to a use case. Prepare a way to compare a simpler workflow with multi-agent orchestration against the required capabilities and interaction flows.
Balancing model quality with data readiness
Likely preparation challenge: benchmarking an LLM or RAG approach may expose weaknesses in source data rather than the model itself. Prepare to distinguish retrieval and enrichment changes from cases that warrant fine-tuning or retraining.
Preserving reliability beyond a prototype
Likely preparation challenge: an MVP may demonstrate value before its integrations and services are ready to scale. Prepare to describe how you would prioritize evaluation, modular code, and production constraints while communicating remaining limitations.

Practice your answers

Technical practice

How would you decide which parts of an enterprise use case should become agent capabilities?

What to coverSuggested preparation: Explain how you would clarify the business goal with internal or client teams, then define agent intents, capabilities and interaction flows. Discuss where a simpler workflow might be preferable to adding an agent.

How would you design a multi-agent workflow and integrate its models?

What to coverSuggested preparation: Describe how agents would divide responsibilities, pass information and handle an unsuccessful step. Explain the trade-off between specialized agents and the complexity of orchestrating them.

How would you choose between prompting, fine-tuning and other model approaches?

What to coverSuggested preparation: Compare LLMs and transformer-based models against the task, available data and required performance. Explain how you would test whether prompt changes suffice before committing to fine-tuning.

How would you design and troubleshoot a RAG pipeline?

What to coverSuggested preparation: Walk through document preparation, embeddings, context-aware retrieval and generation, noting that vector databases are a preferred rather than required qualification. Explain how you would distinguish retrieval failures from generation failures.

How would you benchmark an agent solution and decide whether a model needs retraining?

What to coverSuggested preparation: Define tests tied to the intended use case and compare model and end-to-end workflow results against a baseline. Discuss how you would investigate poor results before deciding between data changes, retrieval changes and retraining.

How would you prepare structured and unstructured data for an AI workflow?

What to coverSuggested preparation: Discuss preprocessing and feature engineering, including when entity extraction, topic modeling or summarization would add useful information. Explain how you would check that enrichment improves the intended model outcome rather than merely adding complexity.

How would you design a data pipeline that supports continuous improvement?

What to coverSuggested preparation: Describe how prepared data and evaluation results could feed later iterations of the solution. Explain how you would detect data-quality changes and assess whether an update actually improves performance.

How would your architecture handle text, audio, video and structured data?

What to coverSuggested preparation: Explain how you would define each modality's input, model interaction and output within an agent workflow. Discuss the trade-off between a shared interaction flow and modality-specific processing.

How would you train and evaluate an ML or deep-learning component for an enterprise application?

What to coverSuggested preparation: Cover task definition, feature choices, training and evaluation data, and comparison with a baseline. Explain how evaluation findings would affect the decision to integrate the component into a larger AI service.

How would you take an AI-agent MVP into a production-grade service?

What to coverSuggested preparation: Describe how you would turn the prototype into modular AI integrations, scalable services and usable interfaces with developers. Explain which assumptions you would validate before scaling and how you would assess the resulting solution.

Behavioural practice

Tell me about a time you turned an unclear stakeholder need into a defined AI use case.

What to coverSuggested preparation: Use a real situation, your consultation actions and the resulting decision or outcome. Show how you clarified goals and determined what the workflow needed to do.

Tell me about a time you explained an AI solution's limitations to a non-technical stakeholder.

What to coverSuggested preparation: Describe the situation, how you communicated the trade-offs and what resulted. Focus on how your explanation supported an actionable decision without overstating capability.

Tell me about a time you owned delivery from prototype toward production.

What to coverSuggested preparation: Give the situation, the decisions and actions you owned, and the result. Distinguish your contribution from the team's work, including how you addressed issues uncovered after the prototype.

Tell me about a time you worked with developers to turn an AI design into an implementation.

What to coverSuggested preparation: Explain the situation, how you agreed on interfaces or responsibilities, and the result. Highlight a concrete trade-off between architectural goals and implementation constraints.

Tell me about a time model or data performance fell short of expectations.

What to coverSuggested preparation: Set out the situation, how you investigated and prioritized possible causes, and the outcome. Describe the evidence behind your next step, whether it involved data, model choice or the workflow.

Sharpen your knowledge

Topics to prepare

Agent workflow and orchestration architecture
The role defines agent capabilities, intents, interaction flows and multi-agent architecture. For revision, sketch a hypothetical client use case, map agent responsibilities and handoffs, and explain why each agent is needed.
LLM, transformer and RAG strategy
The posting calls for selecting, fine-tuning and integrating LLMs, transformer-based models and RAG pipelines. For revision, compare when prompting, retrieval or fine-tuning would be appropriate for one hypothetical use case, including their limitations.
ML evaluation and retraining decisions
Training and evaluating ML models, benchmarking performance and assessing retraining needs are explicit responsibilities. For revision, outline a benchmark with a baseline, representative test cases and criteria for investigating degraded results.
NLP and data intelligence
The role requires structured and unstructured data feature engineering and covers entity extraction, topic modeling and summarization. For revision, trace how each technique could enrich a sample document set and what information might be lost or distorted.
Multimodal model integration
The posting specifies model interactions with text, audio, video and structured data, and requires exposure to multimodal agent delivery. For revision, diagram how those inputs would reach an agent workflow and identify where formats or information quality could affect the result.
Retrieval components
Embeddings, vector databases and context-aware retrieval are preferred qualifications, not stated requirements. As optional preparation, explain how a query could retrieve relevant passages and where irrelevant context could affect an answer.

Skills to revise

Design an agent interaction flow
Practice turning a hypothetical stakeholder request into stated intents, agent capabilities, handoffs and a simple architecture diagram. This mirrors the role’s consultation and workflow-design responsibilities; the scenario is a practice exercise, not a stated Lyzr AI project.
Build and assess a small RAG prototype
As a practice exercise, connect sample documents, retrieval and an LLM response, then check answers against the source passages. This rehearses the posting’s RAG integration and benchmarking work without assuming a required tool or framework.
Prepare and enrich sample data
Practice preprocessing a small mix of structured records and text, extracting entities and producing summaries; record how the transformations affect usable features. This relates to the stated data-preparation and NLP-enrichment responsibilities.
Integrate AI components in modular code
Practice separating data preparation, model calls and evaluation into replaceable components, then explain how a prototype could become a scalable service. The posting explicitly asks for modular integration code and progression from MVP to production-grade solutions.
Present an architecture decision to stakeholders
Practice a short non-technical explanation of a hypothetical agent design, its expected capabilities, limitations and evaluation plan. This reflects the role’s stakeholder consultation, product advice and end-to-end delivery ownership.

Your preparation checklist

Map the stated requirements to evidence
The posting specifies 8+ years overall, with experience in ML, NLP, and LLMs. Prepare a brief evidence sheet with accurate dates, your contributions, and examples of model training, evaluation, enterprise applications, and agent work; mark any gaps plainly.
Prepare an end-to-end project walkthrough
The role spans consultation, architecture, implementation, and delivery. Choose a real project, if available, and prepare a diagram and concise account of the use case, your decisions, modular integration code, collaborators, measured results, and limitations.
Practice an agent-workflow design exercise
For a hypothetical enterprise use case, sketch agent capabilities, intents, interaction flows, orchestration, and model interfaces. State assumptions and explain when multiple agents would—or would not—be warranted.
Build a model-strategy comparison
The posting calls for LLM, RAG, and transformer selection, integration, benchmarking, and assessment of fine-tuning or retraining needs. Prepare a comparison with evaluation measures, test cases, trade-offs, and a reasoned recommendation.
Outline the data and modality plan
Draft a sample pipeline for structured and unstructured data, including preprocessing, feature engineering, relevant NLP enrichment, and a feedback path for improvement. Show how you would assess text, audio, or video only where the use case calls for them.
Explain an MVP-to-production path
Prepare a small architecture or code sample showing modular AI-component integration and scalable service interfaces. Be ready to explain what you would change between a prototype and a production-grade solution; cloud platforms and orchestration frameworks are preferred, not stated as mandatory.
Rehearse stakeholder explanations and questions
Practice a non-technical explanation of your design, its capabilities, limitations, and trade-offs. Prepare questions about the initial use cases, available data, evaluation expectations, and how architecture and delivery responsibilities are shared with client and developer teams.

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…