Shubham kumar profile photo 1 on Tal

Shubham kumar

Immediate Joiner| AI Engineer@ Deccan AI | Python | AI Training Specialist (RLHF & SFT) Operations | SQL | Automating Data Pipelines with Python & Git | LLM Training Data Quality

Mushahari

4+ years of work experience

Deccan AI
Feb 20267 mosPresent

Hyderabad

Improved AI model accuracy and safety by executing high-quality SFT and RLHF workflows, ensuring the model's responses are aligned with human intent and factual correctness. Managed end-to-end Data Validation Pipelines, including Fact-checking (Oracle), Data Deduplication, and Contamination Analysis to ensure high-fidelity training sets for LLMs. Evaluated and curated complex datasets for LLM training, ensuring high accuracy, factual consistency, and adherence to safety guidelines. Designed and automated data pipelines using SQL, reducing manual effort and improving the speed of data delivery for AI model training.
Cognizant
Dec 20241 yr 9 mosPresent

Kolkata, West Bengal, India

Designed and developed a web application for a US-based client in the Travel and leisure industry, aimed at improving hotel management and supporting business growth. Engaged in development activities to enhance various Spring and Springboot applications. Worked as a Backend developer to create REST API's with spring boot framework. Requirements analysis, design and development or various features and enhancement. Knowledge of application and software for tracking problems, such as JIRA. Participated in the development and deployment of migration techniques for a variety of applications. Created the approach and mapping documents about the requirements which help people across the teams to understand it better.
Cognizant
Dec 20231 yr 7 mosJun 2025

Kolkata, West Bengal, India

Currently, I am working in Gen-AI(Docusnap) project as a prompt engineer and python (backend) In the first stage I did research and development on multi-query and HYDE, after that wrote the code and implemented it and then checked whether the response is coming or not. In the second phase I defined how the web based loader loads the content from a specified link or from a specified PDF. In the third phase, first create a POC using fast api, create an endpoint which uses websocket for connection and connect to postman, then implement the same POC in a multiquery application and then check that response is generated from postman. It's like I asked a question, then based on that question it will generate 4 more questions and answer only query which asked in the body of postman in json format. I have created a POC using fast-api, websocket in which response should be come in streaming forms in chunking manner.
Cognizant
Dec 20221 yr 1 moDec 2023

Pune, Maharashtra, India

I was working in this project Gen-AI(Agile 360):- "Agile 360" is a part of" Gen-AI project". In first phase we have to create a "prompt" for creating a jira story in json format using ChatGPT and then create a demo chatbot. After completion of first phase, I am refining the"prompt implementation" inside the project. -> Our requirement is that the result comes from OpenAI will be provided by GPT, and it's up to us how accurate the result will be. We keep modifying it accordingly. a. Developing, testing and refining AI-generated text prompts. b.Collaborating with content, product and data teams to align prompts with company goals and user needs. c.Continuously improving prompt quality, performance and the overall AI prompt generation process. We have also used context setting for fetching data. d. I have added multiple requirements in "Prompt" to create jira issues, also checking and inspecting its response and extracted data, also creating multiple config files for Coversational Business Analyst Al under Generative Al project. In this LLM models are used. e. I have also added code generation prompt and checked it multiple times and for data summarization part( In data summarization, I had to check whether the descriptions matched from key points which is given by the summary). f. I have implemented code generation part using python code from which frontend, backend and database code forms in a list of string using split method.
Cognizant
Feb 20226 mosJul 2022

*Hands-on training* . Core Java . DBMS . Spring boot . React .Microservices *Project* Technologies:- React, html, css(frontend) Springboot using java, RestApi( backend). Tourism management system is an integrated software developed for tourism business.It is a dynamic and responsive system and it addresses the challenging of managing the hotels, addition of hotels, views and updates of hotels, views and add packages of hotels, user booking history admin can see , feedbacks from user and admin both sides and finally logout system is also available.
GitHub
@Kumarshubham123

Shubham’s AI Stack

claude
claude
gemini
gemini
hugging face
hugging face
langchain
langchain
langgraph
langgraph
langsmith
langsmith

Education

Birla Institute of Technology, Mesra

Master of Computer Applications - MCA, Computer Science

Birla Institute of Technology, Mesra

20193 yrs2022
Birla Institute of Technology, Mesra

Bachelor of computer applications -BCA, Computer Science

Birla Institute of Technology, Mesra

20153 yrs2018

Join Shubham and 10k+ other professionals on

Tal Boss

Tal is the app where bosses meet engineers & builders directly, without middlemen

Download Tal Boss

Shubham kumar on Tal

Headline

Immediate Joiner| AI Engineer@ Deccan AI | Python | AI Training Specialist (RLHF & SFT) Operations | SQL | Automating Data Pipelines with Python & Git | LLM Training Data Quality

About

An innovative person with ability to adapt in all situations and conditions. Possess great analytical, logical & interpersonal skills for the enhancement of the organization as well as personal growth. And thanks for visiting my profile 😊.

Location

Mushahari

AI stack

claude

gemini

hugging face

langchain

langgraph

langsmith

Work experience

Generative AI Engineer at Deccan AI, Feb 2026 - Present · 7 mos, Hyderabad, Improved AI model accuracy and safety by executing high-quality SFT and RLHF workflows, ensuring the model's responses are aligned with human intent and factual correctness. Managed end-to-end Data Validation Pipelines, including Fact-checking (Oracle), Data Deduplication, and Contamination Analysis to ensure high-fidelity training sets for LLMs. Evaluated and curated complex datasets for LLM training, ensuring high accuracy, factual consistency, and adherence to safety guidelines. Designed and automated data pipelines using SQL, reducing manual effort and improving the speed of data delivery for AI model training.

Software Engineer at Cognizant, Dec 2024 - Present · 1 yr 9 mos, Kolkata, West Bengal, India, Designed and developed a web application for a US-based client in the Travel and leisure industry, aimed at improving hotel management and supporting business growth. Engaged in development activities to enhance various Spring and Springboot applications. Worked as a Backend developer to create REST API's with spring boot framework. Requirements analysis, design and development or various features and enhancement. Knowledge of application and software for tracking problems, such as JIRA. Participated in the development and deployment of migration techniques for a variety of applications. Created the approach and mapping documents about the requirements which help people across the teams to understand it better.

Junior Software Engineer at Cognizant, Dec 2023 - Jun 2025 · 1 yr 7 mos, Kolkata, West Bengal, India, Currently, I am working in Gen-AI(Docusnap) project as a prompt engineer and python (backend) In the first stage I did research and development on multi-query and HYDE, after that wrote the code and implemented it and then checked whether the response is coming or not. In the second phase I defined how the web based loader loads the content from a specified link or from a specified PDF. In the third phase, first create a POC using fast api, create an endpoint which uses websocket for connection and connect to postman, then implement the same POC in a multiquery application and then check that response is generated from postman. It's like I asked a question, then based on that question it will generate 4 more questions and answer only query which asked in the body of postman in json format. I have created a POC using fast-api, websocket in which response should be come in streaming forms in chunking manner.

Programmer Analyst Trainee at Cognizant, Dec 2022 - Dec 2023 · 1 yr 1 mo, Pune, Maharashtra, India, I was working in this project Gen-AI(Agile 360):- "Agile 360" is a part of" Gen-AI project". In first phase we have to create a "prompt" for creating a jira story in json format using ChatGPT and then create a demo chatbot. After completion of first phase, I am refining the"prompt implementation" inside the project. -> Our requirement is that the result comes from OpenAI will be provided by GPT, and it's up to us how accurate the result will be. We keep modifying it accordingly. a. Developing, testing and refining AI-generated text prompts. b.Collaborating with content, product and data teams to align prompts with company goals and user needs. c.Continuously improving prompt quality, performance and the overall AI prompt generation process. We have also used context setting for fetching data. d. I have added multiple requirements in "Prompt" to create jira issues, also checking and inspecting its response and extracted data, also creating multiple config files for Coversational Business Analyst Al under Generative Al project. In this LLM models are used. e. I have also added code generation prompt and checked it multiple times and for data summarization part( In data summarization, I had to check whether the descriptions matched from key points which is given by the summary). f. I have implemented code generation part using python code from which frontend, backend and database code forms in a list of string using split method.

Programmer Analyst Trainee at Cognizant, Feb 2022 - Jul 2022 · 6 mos, *Hands-on training* . Core Java . DBMS . Spring boot . React .Microservices *Project* Technologies:- React, html, css(frontend) Springboot using java, RestApi( backend). Tourism management system is an integrated software developed for tourism business.It is a dynamic and responsive system and it addresses the challenging of managing the hotels, addition of hotels, views and updates of hotels, views and add packages of hotels, user booking history admin can see , feedbacks from user and admin both sides and finally logout system is also available.

Education

Master of Computer Applications - MCA, Computer Science, Birla Institute of Technology, Mesra

Bachelor of computer applications -BCA, Computer Science, Birla Institute of Technology, Mesra

Public links

github: https://github.com/Kumarshubham123