profile-pic
Vetted Talent

Deepak kumar

Vetted Talent
Passionate and Skilled SDET Engineer with years months of hands-on experience in supporting, automating, and microservice functional, integration testing, API and UI automation and user behaviour analytics using app trackers.
  • Role

    RAG Engineer

  • Years of Experience

    7 years

Skillsets

  • Kubeflow
  • fine-tuning
  • Exploratory data analysis
  • EDA
  • DevOps
  • Datastax Cassandra
  • Databricks
  • Apache Airflow
  • MLFlow
  • Kubernetes
  • Geval
  • Docker
  • Deepseek ocr
  • containerisation
  • Computer Vision
  • Azure
  • AWS
  • Toxicity testing
  • Tableau
  • PyTorch Lightning
  • Yolo5
  • Yolo
  • Transformers
  • SQL
  • RoBERTa
  • Retrieval-Augmented Generation
  • REST Assured
  • rag
  • Multi-Agent Systems
  • Pinecone
  • OpenAI
  • Neo4j
  • LLMs
  • Large Language Models
  • Karate
  • Hugging Face
  • Giskard
  • Sagemaker
  • Python
  • Scikit-learn
  • Prompt Engineering
  • pandas
  • NumPy
  • Matplotlib
  • Feature Engineering
  • CUDA
  • AWS Lambda
  • FastAPI
  • OpenCV
  • MLOps
  • LangChain - 2 Years
  • FastAPI
  • CI/CD
  • CI/CD
  • PyTorch - 2 Years
  • Python - 3 Years
  • Deep Learning
  • Model reliability
  • model optimization
  • Model evaluation
  • Model deployment
  • ML Pipeline
  • LangGraph
  • Hallucination testing
  • foundation models
  • Java - 5 Years
  • Bias testing
  • benchmarking
  • Artificial Intelligence
  • AI Agents
  • Machine Learning
  • LangChain
  • AWS Lambda
  • Generative AI

Vetted For

3Skills
  • Roles & Skills
  • Results
  • Details
  • icon-skill_image
    Senior QA AnalystAI Screening
  • 50%
    icon-arrow-down
  • Skills assessed :Quality Analyst, Automation Testing, Quality Assurance
  • Score: 50/100

Professional Summary

7Years
  • May, 2025 - Present 10 months

    AI ML Engineer

    Sony
  • Jun, 2024 - Mar, 2025 9 months

    Generative AI Engineer

    BCT Consultancy
  • Feb, 2023 - Jun, 20241 yr 4 months

    Data Scientist

    Cred
  • Jun, 2018 - Apr, 20212 yr 10 months

    SDET

    Coviam Technologies
  • Apr, 2021 - Dec, 2021 8 months

    SDET

    Byju's Think & Learn
  • Dec, 2021 - Feb, 20231 yr 2 months

    Software Development Engineer in Test (SDET) AI

    Navi Technology

Applications & Tools Known

  • icon-tool

    Appium

  • icon-tool

    Selenium

  • icon-tool

    REST API

  • icon-tool

    REST Assured

  • icon-tool

    Deep Learning

  • icon-tool

    Java

  • icon-tool

    Python

  • icon-tool

    pytest

  • icon-tool

    testng

  • icon-tool

    Cucumber

  • icon-tool

    Rest Assured

  • icon-tool

    Appium

  • icon-tool

    Espresso

  • icon-tool

    Cypress

  • icon-tool

    Jmeter

  • icon-tool

    Testng

  • icon-tool

    Cucumber

  • icon-tool

    Jenkins

  • icon-tool

    Github

  • icon-tool

    Bedrock

  • icon-tool

    Llama

  • icon-tool

    Jira

  • icon-tool

    MongoDB

  • icon-tool

    Robo3T

  • icon-tool

    Spring Boot

  • icon-tool

    Android Studio

  • icon-tool

    PostgreSQL

  • icon-tool

    LLM

  • icon-tool

    CNN

  • icon-tool

    RNN

  • icon-tool

    YOLO

  • icon-tool

    Postman

  • icon-tool

    Hadoop

  • icon-tool

    Apache Spark

  • icon-tool

    Tableau

  • icon-tool

    OpenCV

  • icon-tool

    Agile Methodologies

  • icon-tool

    Firebase

  • icon-tool

    Crashlytics

  • icon-tool

    Springboot

  • icon-tool

    Grafana

  • icon-tool

    Kibana

  • icon-tool

    Kafka

  • icon-tool

    Kubernetes

  • icon-tool

    Report Portal

  • icon-tool

    Testing

  • icon-tool

    Maven

  • icon-tool

    AWS

  • icon-tool

    Redis

  • icon-tool

    AWS Sagemaker

  • icon-tool

    AWS Bedrock

  • icon-tool

    Tensorflow

  • icon-tool

    Pinecone

  • icon-tool

    FastAPI

  • icon-tool

    LLM

  • icon-tool

    CNN

  • icon-tool

    FAISS

  • icon-tool

    Tableau

  • icon-tool

    Maven

Work History

7Years

AI ML Engineer

Sony
May, 2025 - Present 10 months
    Established scalable MLOps framework using Kubeflow, Docker, and Kubernetes, enabling repeatable training and production deployment with automated approvals and monitoring. Designed, developed and deployed machine learning models for Sony playstation for the business requirements and worked on the playstation recommendation system store. Created end to end pipeline for models development, hyperparameter tuning deployment and monitoring of the ML models using MLFlow and KubeFlow, Automating preprocessing tasks using Apache Airflow for seamless job orchestration. Fine Tuned Claude Sonnet 4.5 models using q-LORA with internal marketing data to increase the efficiency of QnA system by 7%. Integrated MLflow tracking and CI/CD gating, enforcing reproducibility and model version control; reduced development-to-production cycle time by 70%. Designed Agentic services using A2A (GOOGLES AGENT TO AGENT PROTOCOL Protocol) and MCP (Model Context Protocol), enabling scalable multi-agent communication across gaming platforms considering best LLMOPs practices. Developed Tone Guard, a GenAI-powered chatbot filter transforming abusive inputs into respectful responses, reducing toxicity reports by 40%. Implemented an evaluation pipeline for multi-agent systems using CREW, improving reliability of orchestration and reducing error propagation by 25%. Created classical ML models for PlayStation security and network traffic analysis, detecting anomalies and fraudulent activity with 92% accuracy. Used knowledge graph to increase the accuracy of the dashboard query system where user can query from dashboards which has table charts and maps of the movies produced by Sony Pictures. SQL agents for multi databases query system using knowledge graphs and agentic training on neo4j databases. Evaluated the accuracy to 95% and added human in loop to confirm. Developed machine learning models to production for playstation stores services for recommendation, insights and business requirements to production and monitored the performance and drifts. Used MLFlow and databricks and sagemaker for scalable and versioning of the model infrastructure. Did POC on voice agents using Whisper, Langchain and Langgraph to convert a human voice to a characters voice like Ironman and Spiderman.

Generative AI Engineer

BCT Consultancy
Jun, 2024 - Mar, 2025 9 months
    Managing end-to-end model lifecycle, steering problem scoping, data preparation, experimentation, deployment and scaling. Developing and deploying Gen AI applications and multi-level AI agents using OpenAI, Claude and AWS Bedrock Agents, accelerating data access and querying. FineTuned GPT 3.5 to increase the accuracy of the Scientific E Auth system (QnA) system by 11%. Implemented ROUGE based and LLM as judge based accuracy parameter to evaluate generation content. Creating a Q&A system using RAG, Pinecone and Roberta, facilitating query and summarization of drug release documents. Exploring Hugging Face LLMs, recommending appropriate models for query management. Improved retrieval speed by 20% using query translation (e.g., RAG Fusion), indexing, re-ranking, and caching techniques. Conducting systematic experimentation and hyperparameter optimization, enhancing model performance. Optimizing the foundation models using LangChain, enhancing response quality. Conducting exploratory data analysis (EDA), feature engineering and statistical modelling to drive actionable insights for ML systems. Led POC and migrated agents to MCP and A2A methods to streamline LLM data access. Building Cortex Service to enable prompt engineering with LLMs via FastAPI. Securing company data using Agentic AI and supporting AI agents with LangGraph-based orchestration. Designing evaluation pipelines for GenAI summarization with GEval and Giskard, for monitoring LLM performance. Maintaining production-grade DataStax Cassandra DB, optimizing RAG system performance and data retrieval speeds. Incorporating MLOps/LLMOps pipelines for Gen AI projects, streamlining project workflow and delivery. Validating LLMs for hallucinations, biases and toxicity with GLUE, Giskard and RAGAS, improving model safety and reliability. Customizing test sets and metrics to validate factual accuracy, ethical alignment, and output reliability across diverse prompts.

Data Scientist

Cred
Feb, 2023 - Jun, 20241 yr 4 months
    Developed evaluation matrices for AI-based lending approval products, testing and improving product performance. Optimized ChromaDb and PineCone integration, driving LLM retrieval systems to reduce query latency and ensure real-time response. Built test cases and Gen AI evaluation frameworks using RAGAS, GEVal and Giskard, automating and improving testing throughput by 40%. Created chatbots for internal document queries, for accessing data using natural language by the OPS team. Steered ML pipeline development and CI/CD integration, enhancing workflow efficiency. Optimized LLMs for enterprise and customer-facing applications, reducing hallucination and enhancing performance. Engineered LLM fine-tuning workflows using Hugging Face, PyTorch Lightning and Sagemaker, accelerating training and deployment. Implemented computer vision models with OpenCV, Tensorflow, Yolo5, AWS Sagemaker and AWS Lambda to identify and report warehouse product defects. Deployed AI models on AWS bedrock, aligning cross-functional business objectives with technical feasibility.

Software Development Engineer in Test (SDET) AI

Navi Technology
Dec, 2021 - Feb, 20231 yr 2 months
    Developed machine learning models and AI chatbots using natural language processing (NLP) methods and data science library like pandas, scikit learn and tensorflow. Trained, evaluated, and optimized models using machine learning and statistical techniques. Analyzed large-scale datasets, identifying discrepancies and improving model accuracy and performance. Evaluated and improved ML model and chatbot performance, providing insightful feedback. Created AI chatbot test automation framework, enhancing test efficiency and defect detection. Automated API and UI testing, optimizing software quality, performance and release time.

SDET

Byju's Think & Learn
Apr, 2021 - Dec, 2021 8 months
    Managed software testing lifecycle (STLC) and quality processes, delivering Web and mobile applications. Created test automation frameworks and scripts using Selenium, Rest Assured and open-source tools, driving efficient releases. Simulated test cases, tracking test holes and refining test plans against code coverage data. Automated regression suite execution with Jenkins, reducing development cycle and release time. Enhanced system efficiency and reliability with automated data tracking from mobile and web applications.

SDET

Coviam Technologies
Jun, 2018 - Apr, 20212 yr 10 months
    Conducted manual and automation testing of AI chatbots, identifying and reporting bugs to the development team. Curtailed manual testing efforts by 40%, implementing automated test methods and tools. Tracked, analyzed and documented test issues, risks and solutions, facilitating product progress and on-time release. Facilitated continuous product improvement via thorough code review and constructive feedback. Mentored junior engineers on industry best practices, enhancing productivity and software quality.

Major Projects

4Projects

OPS PORTAL

NAVI TECHNOLOGIES
Mar, 2022 - Dec, 20231 yr 9 months

    Worked on the qa for ops portal and led the automation team for regression of services

Mobile APi

Quinnbay Technologies/Coviam Technologies
Jun, 2018 - Jan, 20234 yr 7 months

    Worked in a Mobile team and its backend service

Gamification

Coviam Technologies
Dec, 2019 - Apr, 20211 yr 4 months

    Woked on qa and automation for game service from scratc.

ecting mobile API service and app migration to di erent rest API service (cid:123)

Education

  • BTech: Electronics & Communication Engineering

    Maharaja Agrasen Institute of Technology

Certifications

  • Deep Learning Specialzation

    Coursera

AI-interview Questions & Answers

Basically, I work at grid. Currently, I'm working at grid. I have, like, total 5 in both back end as well as mobile as well as web testing. I started my career at Kinvey Technology, uh, formerly known as Covium Technology. So I started as an a manual over there, and then gradually move to, uh, like, uh, like, uh, gradually converted to software development engineering test. Uh, so, basically, my day day to day job over there was, uh, like, testing both mobile as well as back end, back end APIs. Uh, I used to handle some of the back end ports over there. Then I moved to, um, Baidu's technology where I was a, um, mobile as well as web automation engineer over there, and I learned mobile mobile as well as web automation over there. Then gradually, uh, after after that, I moved to Navi Technologies in 2021 at December, and I worked as a lead as the SDET over there. So after in in in February 2023, I moved to great technology, which I'm currently working over here. So, basically, I'm currently, I'm leading a port, which is a win port. So this is it this is a port for playing games inside the CRED app, and we, like, give rewards to the user who who, like, whoever wins inside the game. So, basically, my day to day job over is handling the port features, all the features that are coming in the sprint and as well as automating all those scenarios, uh, for mobile app as well as as as well as

Uh, we we have used, uh, Jira technology. So we used to create tickets, uh, as well as we I have used, uh, like, uh, Confluence mostly also. So we used to create tickets, and those tickets used to go to develop, uh, different development, uh, cycles, like, uh, different site process cycles, like development to the product manager and then finally coming to the QA, and we used to close those tickets.

Black box texting, uh, testing is generally we can say that, uh, the when you don't know the context of the code, like, uh, if whatever the if you know what is the code, and then it is called white box text. And so, basically, if you are doing, let's say, q unit testing, it is a white box texting because, uh, when you need testing, you are annoying what the code is inside the development inside the development code. Uh, it is done generally in inside the development code, while this white book text texting. Coming to the black book texting, we consider that the regression testing, smoke testing and what it is called sanity testing. These kind of testing are like a black box testing because we don't know what the code is inside the double code is there. So, uh, these are black box

So, basically, I have worked with product managers as well as development team. So, basically, uh, we used to grasp all the testing scenarios, what output of the product is and how it is gonna, like, uh, influence the customer as well as our business values. Then you we used to write the test cost cases, and you'll be used to Get it reviewed by developers and as well as product managers. And after the after this, we used to constantly check Or, like, uh, we used to give some improvements, uh, in the product also, like, uh, or knowing the product Product architecture, how it can be faster, how it can be more better, uh, when it is coming to, you know, use users You just perfect perspective. Sometimes you used you used to get, uh, like, missed by the product managers and developers. So we used to add some values over there. So, basically, uh, like, I can say that, uh, I have mostly worked with, uh, like, product managers teams and as well as, uh, developer teams, mostly with them with them and used to, like, work on the whole release cycle as well as after um, even after the release has happened, we used to work uh

The Okay. I'm not aware of this Industry standards. Um, what is

Script language, I used Python before this. We used BDD for making the framework inside for Python And we used Selenium and all in Python only. When it is coming to the script language, I am also well aware in Java. So we also use TestNG as well as Java, uh, and Cucumber, and let's sorry. Or in Selenium, Nappy, and or test Java. So Python, when it is coming to script language, I use Python mostly.

Uh, basically, uh, if the all the p naught, uh, p 1 and p 2 are covered, uh, and user is not blocked. The when it is coming to the product, user is able to use the product, uh, then we can say that, uh, our product meets the regulatory compliance standard. So, basically, uh, in our current company, if the UAT, your acceptance users, acceptance test scenarios as well as some p p ones are meet met, then we are, like, we are okay with their compliance standards.

From scratch. So, well, it depends upon what type of, uh, what type of, uh, the framework we are developing. If if it is a back end only, so we will better approach it by, like, uh, how many services are being called inside this, uh, like, how many services are being called and how many teams are uh, working collaboratively, uh, inside the this framework. So the these 2 things are they're, like, most basic things, and how many API calls, uh, like, how many services and do we need database access or how will be yeah. Like, we'll be evaluating the uh, truth, uh, truth cases, like, uh, if something is coming from the response, how is it how is it expected to be? Uh, so, basically, we will try to model modularize all these things inside our, uh, inside our back end inside our back end test framework. So basically, uh, when it is coming to UI, when it comes to UI, uh, let's say mobile, we will develop a try to develop a framework where it is, like, uh, both mobile and UI can be handled inside in the same framework. So, uh, we can create, uh, like, interface, uh, which can be, uh, like, which can be implemented by both mobile as well, web, and all the scenarios, which are same in mobile. And web uh, should have different methods when come come when it implementation method, when it comes to, uh, when what do you say? From web and mobile. So, basically, all the implementations will be the test cases will be same, but the implementations will be uh, for both the things. So, basically, we try to, like, uh, develop a framework, which is

Security. So, basically, uh, in terms of QA, Basically, the most important thing is, like, learning what's happening around the globe in terms of QA. There are many such things, like, in in AI also, we need, um, a lot of QA. Uh, and now there has been, like, a very good notion coming over here for AI Testing. So for, like, model testing, what are the models being developed? So big basically, the plans would be, like, we learn and develop because this is a, uh, like, Very new field when it comes to QA. Uh, it, uh, technologically, it is growing very newly. Like, it's a revolution inside currently, it's a revolution inside, um, the way, uh, QA. You can see it. So, basically, uh, we should learn and grow according to the current industry trends. So, basically, we should try to, like, implement I keep trying things, implement things which are coming in the market, and see what

So, basically, uh, role of QA is the, uh, like, is one of the most important, um, role inside the development and the process in improvement. It is because mostly because it ends the, uh, the development prop the product development scenario ends with the QA, and QA are the, like, you know, owner of the product after it is developed and, uh, advanced to the it goes to the users. So, basically, we can say that, uh, like, we are the holder. The QAs in India are the holder of the product. Uh, like, it should it should not be breaking after that it is related. And it generally starts with the business team and product managers, and it ends with the QA only. So we like a QA is has a very important role in product development cycle, and notice the role of QA in engineering process improve. And when it comes to process improvement, uh, QA has a very different very good role because QA interact start from the beginning of, like, uh, when the product is being made and all the portfolios for the product and new designs are being made. Then only, like, QA process, like, QA can, uh, will can try to be mediator between product managers and uh, developers. So that that are time time as well as the efficiency of the process can be mitigated. So, uh, basically, uh, Qwik do handle all the process cycle, and it can miss mitigate a lot of if like, it can increase a lot of if you can see as well, it can mitigate a lot of

Yes. Like, I have I have worked in a load testing for a service which is game a gamification service. So, basically, we used to have a lot of like, we used to have an event called 10, 10, 11, 11, 12, 12 in my first company. So, basically, what we used to service. Like, we you, uh, we used to create events. So during those events, like, uh, even, like, 10,000 of users We are simultaneously playing the games. So we used to check the number of ports that need to be there. Uh, like, it needs to be installed um, for the system. And any parallel ports needs to be installed in this for the system to handle those, um, and those number of, uh, current current users. So, basically, what we how we used to do that, we used to hit, uh, create in inside our, uh, QA environment. What we used to do is, like, Here, the number of ports used we used to check the performance of the, uh, port when we actually used to check the eligibility of the user. Uh, you then used to play the game, And all the process used to be there, uh, making the replay of the actual production environment that used to be inside. In in case when it was, like, in case, uh, when it it was the production, uh, was going on.

I have worked on GitHub workflows by creating, like, automated automated automated, uh, triggering of automation scripts. I have worked on Jenkins as well. I have worked on GoCD also. I don't have much experience in GoCD, but uh, used already configured GoCD pipelines. I have configured GitHub workflows myself only. And I have worked on Jenkins uh, Microsoft only. So, uh, I have created many pipelines in Jenkins using, uh, and I have worked on AWS also.