Looking for a job

Vivek Choudhury

@vivekchoudhury

Technical Architect at Mphasis, architecting Neozeta for COBOL modernization and cutting LLM inference costs by up to 90%.

India

Message

At Mphasis, I designed and delivered Neozeta, an enterprise GenAI platform for automated COBOL and mainframe modernization. It supports Fortune 500 clients and brings together deployments across GCP, AWS, and local environments.

I built a stateful multi-agent orchestration engine with LangGraph-style workflows and MCP tool interfaces. It dispatches agents for business rules, data dictionaries, pseudocode, and Q&A, with recovery and cancellation capabilities.

I also built a provider-agnostic LLM layer for AWS Bedrock, GCP VertexAI, and Azure OpenAI, enabling runtime model switching through YAML configuration. Native prompt caching reduced API inference costs by up to 90%.

Earlier, at EY, I led data science work across ML Ops, retail GenAI, and cybersecurity risk analytics. My experience also includes AML sanction screening, synthetic-image generation, and access-control classification; I’m a co-inventor on three international patent applications.

Experience

Work history, roles, and key accomplishments

Current

Technical Architect / Senior Lead Data Scientist

Current

Mphasis Limited

May 2024 - Present (2 years 5 months)

Designed and delivered Neozeta, Mphasis's flagship enterprise GenAI platform for automated COBOL and mainframe legacy modernization, deployed for Fortune 500 clients. Architected a modular Flask-RESTful backend with 50+ API endpoints, Keycloak OIDC + JWT dual-authentication, Prometheus observability, Docker containerisation, GitLab CI/CD, and PostgreSQL/MySQL with 30+ SQLAlchemy ORM models across

Manager & Risk Officer

Ernst & Young (EY)

Feb 2021 - May 2024 (3 years 3 months)

Led a team of 7 data scientists in designing and deploying LangChain-based Prompt-Response LLM pipelines (GPT-3.5, GPT-4, Google Palm) with HuggingFace FAISS vector embeddings for Rio Tinto. Directed GenAI integration for GCP-hosted retail personalisation for Levi's, and led a 12-member data science team delivering Cyber & Information Security risk analytics for Credit Suisse.

Lang Chain GPT 3.5 GPT 4 Google Palm Hugging Face Faiss AWS Sagemaker Streamlit WhyLabs Helm GCP SQL PySpark Python Anaconda Tableau Anomaly Detection ETL

Senior Data Scientist

ICS

Sep 2020 - Feb 2021 (5 months)

Developed an RNN (LSTM/GRU) and BERT-based AML Sanction Screening classifier for a leading Dubai-based bank, automating High/Low priority classification of NER-generated transaction monitoring alerts. Significantly reduced false positive rates and accelerated Suspicious Transaction Report (STR) generation timelines, improving the bank's compliance posture.

RNN LSTM GRU BERT AML Sanction Screening NER Python Machine Learning

Data Scientist & ML Engineer

Analog Devices via Eximius Design (Wipro)

Aug 2019 - Sep 2020 (1 year 1 month)

Established and maintained AI development and production infrastructure for Analog Devices. Implemented a GAN-based facial recognition system generating high-fidelity synthetic training images using Keras, OpenCV, PyTorch, and TensorFlow; applied Monte Carlo Simulation for system evaluation and hyperparameter tuning.

Gan Keras OpenCV PyTorch TensorFlow Monte Carlo Simulations Embedded Systems Python

Data Scientist

Tata Consultancy Services (TCS)

Jun 2015 - Aug 2019 (4 years 2 months)

Automated enterprise access-control entitlement classification using Linear/Logistic Regression, Random Forest, and K-Nearest Neighbours, integrating trained models into the GETACCESS platform via Python and Flask. Managed large-scale datasets, resolved data quality issues, and optimised Python processing pipelines for improved throughput and system performance.

Linear Regression Logistic Regression Random Forest K Nearest Neighbours Python Flask Data Quality ETL