New opportunity

AI Engineer

Tristone Partners LLP · Mumbai, India

About this role

Tristone Partners LLP is hiring an AI Engineer in Mumbai, India, to embed agentic AI into workflows at its financial services firm. The role focuses on building agents, MCP servers and evaluation processes that make AI-generated code reliable, auditable and ready for production. The engineer will work with senior team members and colleagues across the firm on AI integrations for workflows supporting investment and accounting-related services.

Responsibilities

Design, build and deploy AI-powered tools, agents and workflow automations. Identify repetitive processes that could be automated with AI and workflow orchestration tools.
Work with Claude Cowork, Claude Code, Connectors, Skills and Plugins, alongside platforms such as ChatGPT and Gemini. Evaluate new AI capabilities and tools for potential adoption.
Own the deployment lifecycle, including packaging, testing, hosting and production monitoring. Validate AI-generated outputs for accuracy, completeness and reliability before internal or client-facing use.
Document technical approaches and communicate outcomes to technical and non-technical stakeholders. Follow IT policies, use approved endpoints and enterprise tools, and assess which client or fund data may be sent to each model while maintaining confidentiality and information security.

Requirements

1–3 years of work experience; the posting also specifies 1+ years of relevant experience in AI/ML engineering, software development or applied AI projects. A bachelor’s or master’s degree in computer engineering, computer science or a relevant specialization such as AI/ML, data science or IT is required.
Hands-on experience with AI platforms and prompt engineering; building agentic workflows and integrating external tools through MCP servers or equivalent frameworks; and taking an AI tool or agent from development into production.
A systematic approach to evaluation, including maintained evaluation sets, pre-release regression checks and structured human review. Demonstrable self-built AI projects or agents, Python, SQL, React, Docker, AI coding tools, and clear cross-team communication are required. Projects using Claude tools are particularly valued.

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

Skills for this role

AI agentsAgentic workflowsModel Context Protocol (MCP)MCP serversClaudeClaude CoworkClaude CodeClaude ConnectorsClaude SkillsClaude PluginsChatGPTGeminiCopilotChatGPT CodexPrompt engineeringWorkflow automationWorkflow orchestrationAI evaluationRegression testingHuman reviewPythonSQLReactDockerAPI integrationProduction monitoringTechnical documentationStakeholder communicationData security

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

Build useful agents and integrations
The role designs and deploys agents, MCP servers and workflow automations. Prepare to explain a self-built agent, its tool integrations, and why its workflow design suited the task; Claude-based projects are particularly valued.
Make outputs verifiable
The posting calls for maintained evaluation sets, pre-release regression checks and structured human review—not ad-hoc spot checks. Be ready to describe how you tested accuracy, completeness and reliability before use.
Own the path to production
Responsibilities span packaging, testing, hosting and production monitoring. Prepare a concrete account of taking an AI tool or agent from build to deployment, including how you monitored it.
Protect client and fund information
The role requires judging which data may go to each model, using approved endpoints and enterprise tools, and following IT policies. Prepare to explain how data boundaries shape an AI integration.
Find and communicate practical automations
The engineer works across teams to identify repetitive processes and explain technical approaches to non-technical stakeholders. Prepare to describe how you defined an automation opportunity and communicated its outcome.

Likely challenges to think through

Balancing release speed with dependable results
Preparation hypothesis: quickly shipping AI-generated code may compete with thorough validation and auditability. Practice explaining release criteria, regression checks and when human review should block use.
Using context without exposing sensitive data
Preparation hypothesis: richer client or fund context may improve outputs but restrict which models and endpoints are appropriate. Prepare a decision framework for limiting data, selecting approved tooling and declining unsafe integrations.
Choosing automation over manual work
Preparation hypothesis: a repetitive task may look automatable but still need judgment or exception handling. Prepare to explain how you would identify a bounded use case, measure whether it helps, and retain human review where needed.

Practice your answers

Technical practice

How would you design an agent to automate a repetitive firm workflow?

What to coverSuggested preparation: Explain how you would define the task, choose where the agent can act, and keep a human decision point where needed. Discuss how you would check that the automation improves the workflow without sacrificing accuracy.

How would you integrate an external tool through an MCP server?

What to coverSuggested preparation: Walk through the tool interface, inputs, permissions and handling of failed calls. Explain how you would test the integration and make its actions traceable.

How would you make prompts produce consistent results across models?

What to coverSuggested preparation: Discuss clear task framing, output constraints and iterative testing with representative inputs. Explain what you would recheck when moving between Claude, ChatGPT and Gemini rather than assuming a prompt transfers unchanged.

How would you build an evaluation set and use it before release?

What to coverSuggested preparation: Describe representative cases, expected outcomes and how you would maintain the set as a workflow changes. Explain how regression results would inform a release decision instead of relying on spot checks.

How would you validate AI-generated code and other outputs before use?

What to coverSuggested preparation: Cover testing for correctness, completeness and reliability, plus a structured human review for consequential outputs. Explain how you would record findings so reviewers can understand what was checked and what remains uncertain.

What would you consider when taking an AI tool from build to production?

What to coverSuggested preparation: Walk through packaging with Docker, testing, hosting and production monitoring, all named in the posting. Discuss how you would detect a failure after deployment and decide whether to correct or roll back a change.

How would you use Python and SQL in an AI workflow?

What to coverSuggested preparation: Describe Python’s role in API calls or data preprocessing and SQL’s role in querying structured data. Explain how you would check that data passed between those steps is accurate and appropriate for the task.

What would you include in a simple internal React dashboard for an AI tool?

What to coverSuggested preparation: Focus on information users need to inspect outputs, review status or understand errors; the posting specifies simple internal dashboards, not a particular product. Explain the trade-off between a useful interface and unnecessary complexity.

How would you decide whether client or fund data can be sent to a model?

What to coverSuggested preparation: Start with the posting’s requirements to use approved endpoints and enterprise tooling, follow IT policies and protect confidentiality. Explain when you would withhold data or seek clarification rather than infer that a model is approved.

How would you assess a new AI capability for adoption?

What to coverSuggested preparation: Propose a limited comparison against a defined workflow, considering output quality, integration effort and information-security constraints. Relate the assessment to the role’s use of Claude tools alongside other AI platforms without assuming any specific tool will be adopted.

Behavioural practice

Tell me about a time you identified a process worth automating with others.

What to coverSuggested preparation: Give the situation, how you learned where the manual effort lay, and the actions you took with the affected team. State the result, including any limitation or reason the process was not automated.

Tell me about a time you explained technical work to a non-technical stakeholder.

What to coverSuggested preparation: Describe the situation, what the stakeholder needed to decide and how you adapted your explanation. Give the result and how you checked that your message was understood.

Tell me about a time review or testing changed your planned release.

What to coverSuggested preparation: Set out the situation, the issue found and your actions to investigate or revise the work. Describe the result and what the experience changed about your approach to validation.

Tell me about a time you worked through differing priorities with senior or cross-team colleagues.

What to coverSuggested preparation: Explain the situation and each party’s concern, then describe how you reached a decision or clarified a trade-off. State the result without attributing a decision to yourself if it was shared.

Tell me about a time you handled uncertainty around sensitive information.

What to coverSuggested preparation: Describe the situation, how you established the applicable rules and what action you took before sharing or processing data. Explain the result while keeping confidential details out of your answer.

Sharpen your knowledge

Topics to prepare

Agent workflows and MCP
Suggested revision exercise: Diagram an agent that handles a repetitive accounting-support task, showing where it calls an external tool and where a person approves the result. Posting relevance: The role builds agents, MCP servers or equivalent tool integrations, and workflow automations.
Evaluation and human review
Suggested revision exercise: Define an evaluation set for an AI-generated output, with criteria for accuracy, completeness and reliability, plus a pre-release regression check and review step. Posting relevance: The posting explicitly requires maintained eval sets, regression checks and structured human review.
Client-data boundaries
Suggested revision exercise: For a hypothetical workflow, classify which client or fund fields could be sent to a model and which must be withheld pending approval. Posting relevance: The role requires judgment about model-bound data, approved endpoints, confidentiality and IT policies.
Production lifecycle
Suggested revision exercise: Outline the path from a working agent to packaging, testing, hosting and monitoring; identify what would trigger a release hold. Posting relevance: The engineer is expected to take AI tools and agents into production and monitor them.

Skills to revise

Build a Python tool-using agent
Suggested revision exercise: Build a small agent using synthetic data that calls a Python-backed tool through an MCP server or equivalent framework, then test a failed tool call. Posting relevance: Hands-on agentic workflows, external-tool integration and Python are explicit requirements.
Refine prompts across platforms
Suggested revision exercise: Give the same structured task to two available platforms named in the posting, revise the instructions, and record differences in consistency and output quality. Posting relevance: The role calls for prompt engineering across platforms such as Claude, ChatGPT and Gemini; Claude tools are particularly valued.
Run an evaluation before release
Suggested revision exercise: Create a small, reusable set of synthetic test cases, run it against two agent versions, record regressions and route uncertain outputs for human review. Posting relevance: Systematic validation before internal or client-facing use is required.
Package and observe an agent
Suggested revision exercise: Containerize a small AI tool with Docker, test its startup and failure behavior, and capture basic runtime errors for review. Posting relevance: Docker packaging, testing, hosting and production monitoring appear in the posting.
Query data and display results
Suggested revision exercise: Use SQL to query synthetic structured records, Python to process the result, and React to display a simple internal dashboard. Posting relevance: The requirements name SQL, Python data processing and React dashboards.
Explain an automation to stakeholders
Suggested revision exercise: Write a one-page technical approach and a short non-technical explanation covering the workflow, validation results and data-handling limits. Posting relevance: The role requires documentation and clear communication across teams.

Your preparation checklist

Prepare a project walkthrough
Select a self-built AI agent or workflow and show its purpose, your contribution, tool choices, and results. The posting asks for demonstrable projects and especially values work using Claude tools.
Diagram an agent’s tool integration
Prepare a short architecture sketch showing how an agent invokes an external tool through an MCP server or equivalent framework, including inputs, outputs, and failure handling. Use your own project if possible; otherwise label it as a design exercise.
Build an evaluation example
Bring a small example evaluation set with expected outputs, pre-release regression checks, and a structured human-review rubric. Explain how each would catch inaccurate, incomplete, or unreliable results.
Trace a deployment from build to monitoring
Prepare a deployment walkthrough covering packaging with Docker, testing, hosting, and production monitoring. Identify where Python, SQL, a simple React dashboard, or an AI coding tool fits only if it was part of your work.
Practice a data-handling decision
Work through a hypothetical client or fund data request: explain what you would check before sending data to a model, when you would use approved endpoints and enterprise tools, and when you would seek guidance. The role emphasizes confidentiality and IT policies.
Prepare a cross-team explanation
Draft a brief, non-technical account of one automation opportunity, its limits, and how outputs would be validated before use. This reflects the posting’s emphasis on working across teams and communicating outcomes.
Write questions for the employer
Ask which workflows are priorities, how the team maintains evaluation sets and human review, and what approved tooling and data-handling boundaries apply. Keep these as questions rather than assumptions about current processes.

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…