New opportunity

Applied AI Scientist

FIRST LIVINGSPACES PRIVATE LIMITED · Mumbai, India

About this role

FIRST LIVINGSPACES PRIVATE LIMITED seeks an Applied AI Scientist with 3–7 years of experience to design, prototype, and ship agent-based AI products from concept to production. This hands-on role spans chat agents, voice agents, multilingual AI, workflow automation, and multimodal systems embedded in real products. It is based at the company’s Mumbai office.

Responsibilities

Build and integrate chat, voice, and multimodal agents into production products. Design workflows using LLMs, tools, memory, and state, and build systems in which agents plan, act, and verify outcomes.
Implement voice pipelines covering automatic speech recognition, text-to-speech, turn-taking, and interruptions. Tune Indian-language voice agents for latency and quality.
Develop agent-driven workflow automation, media and marketing-technology agents, and persona-based features for content, campaigns, and engagement.
Implement moderation and control mechanisms; add observability, logging, and debugging; and reduce hallucinations and other failure modes through system design. Collaborate with product, design, and engineering teams.

Required qualifications

A bachelor’s or master’s degree in Computer Science or a related field from a premier institution, plus a strong AI engineering background.
Hands-on experience building LLM-powered applications and chat agents or conversational systems; understanding of tool calling, agent loops, memory, and state; and experience integrating AI APIs and MCPs into backend or product workflows.
Familiarity with voice AI systems, or strong interest in building them; ability to prototype and iterate rapidly, debug real-world AI failures, and work in a fast-moving product environment. The role requires shipping production systems rather than focusing solely on prompts, demos, media creation, or research.

Preferred qualifications include education at IIT, IIIT, ISI, or RECs; experience with voice agents, speech systems, telephony, Indian-language NLP, agent orchestration, RAG, embeddings, vector databases, marketing or campaign automation, personalization, or SaaS products operating at scale. The posting mentions extra paid leave for birthdays and anniversaries.

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

Skills for this role

AI agentsLLMsConversational AIVoice AIMultimodal AIWorkflow automationTool callingAgent orchestrationAgent memoryModel Context Protocol (MCP)AI API integrationAutomatic speech recognition (ASR)Text-to-speech (TTS)Indian language NLPRetrieval-augmented generation (RAG)EmbeddingsVector databasesObservabilityDebuggingHallucination reductionPersonalization

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

Ship agents inside products
The role explicitly spans concept, prototyping, and production integration, rather than prompts or demos alone. Prepare to explain how an AI feature moves from prototype to a working product integration.
Build agents that execute and verify tasks
The posting calls for workflows using LLMs, tools, memory, and state, including agents that plan, act, and verify outcomes. Prepare to describe design choices for tool use, state handling, and checking results.
Develop voice and multilingual experiences
Responsibilities include ASR, TTS, turn-taking, interruptions, and tuning Indian-language voice agents for latency and quality. Prepare to discuss how those components interact and how you would evaluate responsiveness and output quality.
Make AI behavior controllable and debuggable
The role calls for moderation, control mechanisms, observability, logging, debugging, and reducing hallucinations through system design. Prepare to explain how you would detect failures, trace their causes, and limit their effects.

Likely challenges to think through

Balancing voice speed against quality
A likely challenge is keeping turn-taking and interruptions responsive while preserving speech quality across Indian languages; this is a preparation hypothesis, not a reported problem. Prepare to discuss where latency arises in an ASR-to-TTS pipeline and how you would assess trade-offs.
Keeping agent actions reliable
Agents that call tools and retain state may act on incomplete context or fail to verify an outcome. This is an inferred challenge. Prepare to discuss action boundaries, verification steps, and ways to recover from failed tool calls.
Moving quickly without losing control
Rapid iteration and production delivery may compete with time spent on moderation and debugging; the posting does not say this conflict currently exists. Prepare to explain how you would prioritize safeguards and observable behavior while iterating with product, design, and engineering.

Practice your answers

Technical practice

How would you take a chat or multimodal agent from prototype to a production product?

What to coverThe role calls for agents embedded in real products, not standalone demos. Preparation suggestion: explain integration boundaries, how the agent handles inputs and outputs, and what you would validate before and after release.

How would you integrate tool calling, AI APIs, and MCP into an agent workflow?

What to coverThe posting explicitly names tool calling, AI APIs, and MCP integration. Preparation suggestion: discuss tool selection, input validation, error handling, and how to verify that a tool action achieved its intended result.

How would you manage memory and state in a conversational agent?

What to coverThe role calls for workflows using memory and state. Preparation suggestion: distinguish short-lived conversation state from information retained for later use, and explain trade-offs around relevance, correctness, and control.

How would you design a voice-agent pipeline that handles turn-taking and interruptions?

What to coverThe posting names ASR, TTS, turn-taking, and interruptions as parts of the voice pipeline. Preparation suggestion: walk through the interaction from incoming speech to spoken response, including how the system should respond when a user interrupts.

How would you evaluate latency and quality for an Indian-language voice agent?

What to coverThe role specifically calls for tuning Indian-language voice agents for latency and quality. Preparation suggestion: explain how you would test recognition, response, and synthesized speech with representative utterances, then discuss trade-offs between responsiveness and quality.

How would you build an agent-driven workflow that plans, acts, and verifies outcomes?

What to coverThe posting asks for workflow automation and task execution systems with planning, action, and verification. Preparation suggestion: describe how you would define a successful outcome, check each action’s result, and handle a failed or incomplete step.

How would you design a persona-based agent feature for campaigns or engagement?

What to coverThe role includes media and martech agents and persona-based features for content, campaigns, and engagement. Preparation suggestion: explain how a persona would influence outputs while preserving task constraints, and how you would evaluate whether the feature serves its intended product use.

How would you reduce hallucinations and control unsafe or unintended agent behavior?

What to coverThe posting calls for moderation, control mechanisms, and system-level reduction of hallucinations and other failures. Preparation suggestion: discuss checks on generated claims and proposed actions, including when the agent should decline, seek clarification, or avoid taking an action.

How would you observe and debug an AI agent that fails in production?

What to coverThe role explicitly includes observability, logging, and debugging of AI behavior. Preparation suggestion: identify what you would record across model responses, state changes, and tool actions, then explain how you would use those records to locate a failure.

When would you use RAG, embeddings, or a vector database in an agent?

What to coverThese are listed as strong plus points, not required components of every product. Preparation suggestion: explain when retrieval could help an agent use relevant information, and how you would check whether retrieved material actually improves its answers.

Behavioural practice

Tell me about a time you worked with product, design, and engineering on an AI feature.

What to coverThe posting calls for close collaboration across these teams. Use a real situation, the decisions and coordination you handled, and the resulting product outcome or lesson.

Describe a time you rapidly prototyped and revised an AI-driven feature.

What to coverRapid prototyping and iteration are stated expectations. Give a real example covering the initial goal, what feedback or evidence changed your approach, and the result.

Tell me about a difficult real-world AI failure you debugged.

What to coverThe role emphasizes debugging AI behavior beyond demos. Describe the failure, how you investigated it, the action you took, and what the outcome showed.

Describe a time you helped take an AI idea into production.

What to coverThe posting emphasizes shipping systems embedded in products. Use a real example that makes your responsibilities, production-integration decisions, and observable result clear.

Tell me about a time you had to balance an AI feature’s quality with speed of delivery or response.

What to coverThe posting describes a fast-moving product environment and explicitly mentions latency and quality tuning for voice agents. Preparation suggestion: choose a real situation, explain the trade-off and your action, and state the result without claiming it involved voice unless it did.

Sharpen your knowledge

Topics to prepare

Agent loops, memory, and state
Explicit requirement. Suggested revision exercise: diagram a chat agent that plans a task, calls a tool, records state, and verifies the result; explain how it handles a failed tool call.
Voice-agent interaction
The role covers ASR, TTS, turn-taking, interruptions, and Indian-language latency and quality. Suggested revision exercise: trace a voice conversation from speech input to spoken response and identify where delays or interruptions change agent behavior.
Moderation and failure modes
The posting calls for agent controls and reducing hallucinations through system design. Suggested revision exercise: classify several incorrect or unsafe agent responses and choose a control or verification step for each.
Production agent workflows
The role emphasizes AI systems embedded in products rather than standalone demos. Suggested revision exercise: sketch how an agent feature connects to a product workflow, including API integration, logging, and a way to detect failed outcomes.
Retrieval for agent responses
RAG, embeddings, and vector databases are listed as plus points, not requirements. Suggested revision exercise: explain when retrieved context could help an agent answer and how you would check whether it supports the answer.

Skills to revise

Prototype a tool-using chat agent
Suggested hands-on exercise: build a small conversational flow that calls a tool, preserves task state, handles a tool error, and checks the result. This practices the required LLM-application and chat-agent experience.
Integrate an AI API or MCP
Suggested hands-on exercise: connect a prototype agent to an AI API or MCP-backed capability in a backend workflow and inspect a successful and a failed call. The posting explicitly requires integration experience.
Test a voice pipeline
Suggested hands-on exercise: prototype an ASR-to-agent-to-TTS exchange, then test turn-taking and interruption cases while recording latency and response quality. Voice-system familiarity or strong interest is required; deeper voice experience is a plus.
Debug and observe agent behavior
Suggested hands-on exercise: add logs for an agent’s decisions, tool calls, and verification results; use them to diagnose one hallucinated answer and one failed action. Observability and real-world debugging are explicit priorities.
Implement a bounded product workflow
Suggested hands-on exercise: prototype an agent-driven campaign or engagement task with a defined action and outcome check. Media and martech agents, persona-based features, and workflow automation are stated responsibilities.

Your preparation checklist

Map the stated requirements to evidence
Prepare a one-page inventory of relevant experience, education, shipped LLM applications, conversational systems, AI API and MCP integrations, and voice familiarity or interest. Mark gaps rather than inventing examples.
Prepare a production project walkthrough
Select one real project, if available, and outline the problem, your contribution, architecture, prototype-to-release decisions, integration, and outcomes. The posting emphasizes shipped product systems over prompts or demos alone.
Practice an agent workflow design
For a hypothetical product task, draw an LLM agent loop showing tool calls, memory, state, planning, action, and outcome verification. Label it as a preparation exercise, not company architecture.
Sketch a voice-agent pipeline
Prepare a diagram covering ASR, TTS, turn-taking, and interruptions, with proposed ways to measure latency and quality for an Indian-language interaction. Voice-system familiarity or strong interest is required; speech experience is a plus.
Show how you would diagnose failures
Draft a sample trace and evaluation plan for a hallucination or failed tool action. Include logging, observability, moderation or controls, debugging steps, and a way to verify that a fix works.
Outline a product-feature exercise
Sketch a hypothetical agent-driven campaign or engagement workflow with a persona, tool boundaries, and a handoff to the product. The posting names media, martech, and persona-based features; do not present the exercise as an existing company project.
Write focused employer questions
Bring questions about which agent features are closest to production, how voice latency and quality are evaluated, and how product, design, and engineering divide ownership of releases and failures.

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…