New opportunity

Python AI Developer

Kumaran Systems · Chennai, India

About this role

Kumaran Systems seeks a Python AI Developer in Chennai, India, to build data-driven applications, RESTful APIs, and production-grade machine learning and LLM features. The job information lists 4–5 years of work experience; the role summary asks for 3+ years of hands-on experience. The position is in the Banking department and is full-time.

Responsibilities

Design and develop scalable Python services and RESTful APIs using Flask.
Build data-processing pipelines and analytics workflows with Pandas and NumPy.
Implement, train, optimize, and deploy machine learning models. Apply GenAI and LLM techniques to real-world use cases, evaluate outputs, and iterate.
Write and optimize SQL queries and database schemas for Oracle, MySQL, or PostgreSQL.
Maintain code quality through testing, logging, and CI-friendly structures. Work with product and data teams to turn requirements into scalable solutions.

Requirements

Strong Python programming and scripting skills; experience with Pandas, NumPy, SQL, and Flask-based REST API development.
Practical experience with machine learning libraries, preferably PyTorch, and hands-on LLM work involving prompt engineering and integration. The posting cites GPT, Llama2, and Gemini as examples.
Proficiency with at least one of Oracle, MySQL, or PostgreSQL. pytest is cited as an example testing tool.

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

Skills for this role

PythonPandasNumPySQLFlaskREST APIsPyTorchGPTLlama 2GeminiPrompt engineeringOracleMySQLPostgreSQLpytestCIData processing pipelinesMachine learningModel deployment

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

Production-ready Python services
Source-backed: Design robust, scalable Flask REST APIs, with testing, logging, and CI-friendly code. Collaborate with product and data teams to turn requirements into working services.
Data processing and database performance
Source-backed: Build Pandas and NumPy pipelines and analytics workflows; write efficient SQL and optimize queries and schemas in Oracle, MySQL, or PostgreSQL.
Machine learning through deployment
Source-backed: Implement, train, and optimize ML models, preferably using PyTorch, and integrate them into production.
Practical LLM features
Source-backed: Use prompt engineering and LLM integration for real-world use cases, evaluate outputs, and iterate. GPT, Llama2, and Gemini are examples in the posting.

Likely challenges to think through

Balancing model behavior with service reliability
Preparation hypothesis: Putting ML behind a scalable API may require trade-offs among response time, resource use, and model quality. Prepare to explain how you would measure those factors and test and log the service.
Finding bottlenecks across data and SQL
Preparation hypothesis: Pipeline performance may depend on both in-memory processing and database queries. Prepare to walk through how you would locate a bottleneck, preserve correct results, and choose where to optimize.
Evaluating variable LLM outputs
Preparation hypothesis: Prompt changes can improve some outputs while worsening others. Prepare an approach for defining useful evaluation cases, comparing outputs, and deciding when to iterate or retain the current integration.

Practice your answers

Technical practice

How would you structure a Python service so it remains maintainable as it scales?

What to coverThe role calls for robust, scalable Python services. Preparation suggestion: discuss separation of concerns, error handling, and how you would identify performance bottlenecks before optimizing.

How would you design a Flask REST API that serves a data-driven feature?

What to coverThe posting explicitly calls for Flask-based REST APIs. Preparation suggestion: cover request validation, response and error conventions, and how you would keep the API separate from data-processing or model logic.

How would you build and troubleshoot a Pandas and NumPy data-processing pipeline?

What to coverThe role includes data-processing pipelines and analytics workflows using both libraries. Preparation suggestion: explain how you would handle missing or inconsistent data, verify transformations, and investigate memory or runtime problems.

How would you improve a slow SQL query and its supporting schema?

What to coverThe posting calls for efficient SQL and schema optimization in at least one of Oracle, MySQL, or PostgreSQL. Preparation suggestion: discuss examining the query plan, considering indexes and schema changes, and checking their effect on writes and correctness.

How would you train and optimize an ML model for a practical use case?

What to coverThe responsibilities include implementing, training, and optimizing models, preferably with PyTorch. Preparation suggestion: explain your choice of baseline and evaluation metric, how you would detect overfitting, and what would guide further optimization.

What would you check when integrating a trained model into a production service?

What to coverThe role calls for model integration and deployment within production-grade services. Preparation suggestion: cover input consistency, inference latency, failure handling, and how you would verify that deployed behavior matches evaluation results.

How would you integrate an LLM and develop prompts for an application feature?

What to coverThe posting names hands-on LLM integration and prompt engineering, with GPT, Llama2, and Gemini as examples. Preparation suggestion: discuss task instructions, handling application inputs and outputs, and trade-offs such as latency, cost, and unreliable responses.

How would you evaluate an LLM feature and decide what to change?

What to coverEvaluating LLM outputs and iterating are explicit responsibilities. Preparation suggestion: describe a task-relevant test set, criteria for useful and incorrect responses, and how you would distinguish a prompt issue from an integration issue.

How would you test and diagnose a Flask service that uses data processing or a model?

What to coverThe posting emphasizes testing, logging, and CI-friendly code, with pytest cited as an example. Preparation suggestion: distinguish unit from integration tests, explain what useful logs would capture without exposing sensitive data, and describe how automated checks would catch regressions.

Behavioural practice

Tell me about a time you translated a product or data requirement into a technical solution.

What to coverCollaboration with product and data teams is part of the role. Preparation suggestion: use situation, action, and result to show how you clarified ambiguity, agreed on expected behavior, and assessed the outcome.

Tell me about a time you had to balance scalability with delivery constraints.

What to coverThe posting calls for scalable solutions and production-grade features. Preparation suggestion: use situation, action, and result to explain the trade-offs you identified, what you chose, and how you checked whether the choice worked.

Tell me about a time you investigated a reliability or performance problem in a service or data workflow.

What to coverRobust services, efficient SQL, and logging are relevant to this role. Preparation suggestion: use situation, action, and result to describe the evidence you gathered, the fix you made, and how you verified it.

Tell me about a time evaluation or feedback changed an ML or AI feature you were building.

What to coverThe role includes evaluating LLM outputs, iterating, and working across teams. Preparation suggestion: use situation, action, and result to show what the evidence revealed, how you adapted the feature, and what changed afterward.

Sharpen your knowledge

Topics to prepare

REST API and service design
Posting: Build robust, scalable Python services and RESTful APIs with Flask. Suggested revision: sketch endpoints for a model-prediction service and explain request validation, error responses, and how the service would handle growing traffic.
Data-processing workflow design
Posting: Build pipelines and analytics workflows with Pandas and NumPy. Suggested revision: trace how raw records become model-ready data, identifying missing values, data types, and checks at each stage.
ML training and production integration
Posting: Implement, train, optimize, and integrate ML models into production; PyTorch is preferred. Suggested revision: explain a model’s training, evaluation, optimization, and serving steps, including how you would detect a drop in output quality.
LLM prompting and output evaluation
Posting: Integrate GenAI/LLMs, use prompt engineering, evaluate outputs, and iterate; GPT, Llama2, and Gemini are examples. Suggested revision: define evaluation criteria for a sample LLM task and compare two prompt variants against them.
SQL and database performance
Posting: Write efficient SQL and optimize queries and schemas in Oracle, MySQL, or PostgreSQL. Suggested revision: inspect a sample query and schema, then explain a performance bottleneck and a proposed change.
Production code quality
Posting: Maintain testing, logging, and CI-friendly structures. Suggested revision: outline which API, data-processing, and model-integration behaviors need tests and what useful failures should be logged.

Skills to revise

Build a Flask endpoint in Python
Posting: Python scripting and Flask REST API development are must-have skills. Suggested exercise: implement a small JSON endpoint that validates input, calls a separate processing function, and returns clear success and error responses.
Transform data with Pandas and NumPy
Posting: Both libraries are required for data manipulation and numerical computing. Suggested exercise: clean a small table, calculate a derived numeric feature, and check the output for missing or unexpected values.
Train and assess an ML model
Posting: Practical ML-library experience is required; PyTorch is preferred. Suggested exercise: train a small model, record an evaluation metric, make one optimization change, and explain whether the result improved.
Write and tune SQL
Posting: Proficiency in at least one of Oracle, MySQL, or PostgreSQL is required. Suggested exercise: write a join and aggregation in the RDBMS you know, inspect its query plan, and test a query or schema adjustment.
Integrate and evaluate an LLM call
Posting: Hands-on LLM integration and prompt engineering are required. Suggested exercise: connect a small application function to an LLM you can access, compare prompt outputs on a fixed set of inputs, and record failure cases. This is practice, not a specified employer project.
Test and log a service
Posting: Testing, logging, and CI-friendly structure are responsibilities; pytest is given as an example. Suggested exercise: add tests for valid and invalid API requests, log a processing failure without exposing input data, and run the tests with one repeatable command.

Your preparation checklist

Prepare a Python service walkthrough
The posting calls for robust Flask REST APIs. Prepare a project walkthrough, if available, showing endpoint design, request validation, error handling, and how the service was made scalable.
Map a data pipeline from input to output
Pandas and NumPy workflows are stated responsibilities. Bring a concise example or practice design showing data cleaning, transformations, numerical checks, and the resulting analytics output.
Work through a SQL performance example
The role requires SQL proficiency in at least one of Oracle, MySQL, or PostgreSQL. Prepare a query and schema example that explains a performance issue, the change made, and how it was measured.
Explain an ML model’s path to production
The posting covers training, optimization, and production integration, with PyTorch preferred. Prepare a walkthrough of model choice, evaluation, deployment approach, and monitoring considerations; label any proposed design as hypothetical.
Prepare an LLM evaluation exercise
Hands-on prompt engineering and LLM integration are required. Outline a use case, sample prompts, an integration flow, evaluation criteria, and how you would revise poor outputs; do not assume a particular model is mandatory.
Show how you protect code quality
Testing, logging, and CI-friendly structure are stated responsibilities. Prepare a small Flask or pipeline example with focused tests, useful log events, and an explanation of how it could run in CI.
Bring questions about delivery expectations
Ask how product and data teams define requirements, which banking use cases this role supports, and how API, ML, and LLM work is prioritized. Clarify the posting’s 3+ years in the role summary versus 4–5 years in Job Information.

Availability last checked: 4 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…