Prashant's CV
Prashant Chowdhary
Senior Specialist - AI/ML | Machine Learning
Profile summary
Accomplished Senior Specialist - AI/ML with 10 years of progressive experience in machine learning and artificial intelligence engineering. Track record of developing advanced NLP models, building retrieval-augmented generation workflows, and scaling production systems that reduce latency and improve model accuracy. Dedicated to applying technical expertise to deliver robust enterprise solutions and automation infrastructure.
Career highlights
Generative AI and RAG Workflows: Built LLM and RAG workflows that analyze requirements and user stories to generate test scenarios, test cases, BDD feature files, and QA automation scripts at Qualitest.
LLM Latency and Robustness: Cut end-to-end LLM response latency by 40% and improved production model robustness by 25% through retrieval tuning, asynchronous processing, and multi-cloud deployments at Tata Consultancy Services.
Request Classification and Accuracy: Increased enterprise request classification accuracy by 20% using BERT and LSTM models while embedding SHAP, LIME, and ELI5 into machine learning workflows.
Fraud Risk and Transaction Monitoring: Built an XGBoost risk classification model for suspicious transaction monitoring and compliance workflows, reducing fraudulent transaction risk exposure by 30 percent.
Key skills
Professional experience
Directs artificial intelligence and machine learning initiatives, focusing on advanced workflow development, artificial intelligence systems integration, and production monitoring infrastructure.
- Led Qualitest COCO by building large language model and retrieval-augmented generation workflows that analyze requirements and user stories to generate test scenarios, test cases, behavior-driven development feature files, and quality assurance automation scripts.
- Built an artificial intelligence powered GitHub pull request review system using LangGraph multi-agent workflows, webhook ingestion, asynchronous processing, specialized review agents, and automated inline feedback through GitHub APIs.
- Scaled long-running artificial intelligence workflows by integrating PostgreSQL, Redis, Celery, GitHub APIs, and backend services for asynchronous orchestration and production automation.
- Strengthened production traceability by implementing Langfuse across large language model calls and agent workflows to monitor execution paths, latency, failures, and runtime behavior.
Managed matrixed reporting structures while directing cross-functional engineering and product analysis across natural language processing applications and anti-money laundering monitoring systems, overseeing the delivery of artificial intelligence solutions.
- Developed natural language processing models for product features.
- Engineered artificial intelligence pipelines for enterprise systems.
Directed engineering operations for enterprise generative artificial intelligence initiatives, overseeing workflow execution, model robustness frameworks, and team mentorship across multi-cloud deployments.
- Cut end-to-end LLM response latency by 40% by tuning retrieval depth, context size, workflow execution, and unnecessary model calls in RAG-powered applications.
- Improved production model robustness by 25% through timeout and retry policies, fallback handling, asynchronous processing, observability, reusable CI and CD patterns, MLflow tracking, and multi-cloud deployments.
- Accelerated team delivery speed by 15% by mentoring engineers on RAG architecture, prompt engineering, GenAI workflows, and production ML practices.
Directed machine learning evaluation and explainability initiatives within enterprise products, overseeing natural language processing frameworks and collaborating with stakeholders to align technical deployments with business operational goals.
- Increased enterprise request classification accuracy by 20% using BERT and LSTM models, reducing misrouted requests across service workflows.
- Embedded SHAP, LIME, and ELI5 into machine learning workflows and applied precision, recall, and error analysis checks to improve model transparency.
- Translated model outputs with business stakeholders into operational decisions and workflow changes across enterprise deployments.
- Built an XGBoost risk classification model for suspicious transaction monitoring and compliance workflows, reducing fraudulent transaction risk exposure by 30 percent.
- Executed model and pipeline performance improvements through behavioral and transaction level feature engineering, hyperparameter tuning, scalable feature processing, and streaming workflows.
- Shortened average service-desk turnaround time by 40 percent by designing a machine-learning ticket-routing engine using BERT, TF-IDF, and Word2Vec.
- Automated request prioritization and routing by integrating model predictions with cloud-deployed service-desk workflows and resolution queues.
- Builds automated GitHub pull request review systems using LangGraph multi agent workflows and webhook ingestion.
- Integrates PostgreSQL, Redis, Celery, and GitHub APIs to scale long running artificial intelligence workflows for production automation.
- Increased enterprise request-classification accuracy by 20% using BERT and LSTM models.
- Embedded SHAP, LIME, and ELI5 into machine learning workflows to reduce compliance-review effort.
Projects
Implemented an end-to-end RAG pipeline covering ingestion, preprocessing, chunking, embeddings, Qdrant indexing, semantic retrieval, context construction, and LLM generation with LangChain, OpenAI APIs, and Llama 3. Evaluated retrieval and generation quality using context precision, context recall, faithfulness, answer relevancy, and end-to-end response latency.
LangChain OpenAI APIs Llama 3 QdrantEducation
Computer Science & Engineering
Information Science & Engineering