Ankur Dhuriya profile photo 1 on Tal

Ankur Dhuriya

AI Engineer @ ThoughtWorks | Customizing LLMs for Life Sciences & Procurement | Fine-tuning • RAG • RLHF | AI R&D | MLOps

New Delhi

6+ years of work experience

Thoughtworks
Jul 20251 yr 2 mosPresent

Gurugram

Architected and deployed a compliant, production-grade RAG chatbot for LifeScience Preclinical Safety team using AWS Bedrock, Claude Sonnet, and LangGraph; established end-to-end MLOps pipelines via GitHub Actions and Kubernetes with Weights & Biases for observability, implemented strict ALM protocols through SonarQube security gating, and developed a custom “Golden Dataset” evaluation framework that achieved 95% accuracy in a highly regulated pharmaceutical environment while streamlining access to critical research documentation. Agentic AI & MCP Integration: Architected a unified Copilot assistant using LangGraph and LlamaIndex, implementing MCP-based context sharing to enable seamless A2A integrations across Jira, Confluence, GitHub, and Databricks for real-time enterprise knowledge retrieval. Observability & Evaluation Frameworks: Established robust observability using LangSmith and DeepEval, implementing logging, tracing, and prompt tracking to monitor quality metrics. Achieved 94% accuracy in Biomedical QA via a self-reflective reasoning framework (Qwen3/LLaMA) with structured XML outputs. Performance Optimization: Designed a semantic caching layer with Redis, reducing latency from 6s to <800ms and cutting inference costs by 50% for 10K+ daily queries. Integrated Amazon Bedrock and Azure OpenAI to optimize model selection for cost/performance trade-offs. API & Microservices Development: Developed secure, scalable FastAPI microservices for AI inference, adhering to high-quality Python coding standards (modularity, testing, documentation) and ensuring seamless integration with frontend interfaces.
GEP Worldwide
Dec 20248 mosJul 2025

Hyderabad

ETL & Document Pipelines: Built end-to-end ETL pipelines for document ingestion, parsing (OCR/IDP), and enrichment, processing 50K+ invoices/month. Implemented hybrid retrieval (BM25 + embeddings) over SQL/NoSQL databases and Vector Stores (Pinecone) for accurate RAG-based insights. RAG & LLM Integration: Engineered a spend analytics NLQ chatbot using LangChain and OpenAI/Claude models, achieving 92% answer relevance. Collaborated with SMEs to translate business needs into technical solutions, reducing AP cycle time by 35%. CI/CD & Productionization: Implemented robust CI/CD pipelines using Azure DevOps for automated testing and deployment of AI models to Azure Kubernetes Service (AKS), ensuring reliable production releases and version control for prompts and code. Security & Governance: Integrated custom guardrails for PII detection and prompt injection filtering, ensuring secure and compliant AI interactions in enterprise environments.
Builder.ai
Dec 20222 yrs 1 moDec 2024

Gurugram

time by 70% (45min → 12min) with 94% syntactic correctness. Microservices & API Design: Designed and maintained robust FastAPI microservices for mul- tilingual transcription summarization, processing 10K+ calls/day across 8 languages. Ensured high-quality Python code with comprehensive unit tests and modular design. Observability & Monitoring: Implemented comprehensive logging and tracing for agent runs, enabling rapid debugging and performance tuning. Automated evaluation pipelines tracked halluci- nation rates and safety scores (99.2%), ensuring system reliability. Cloud Deployment: Deployed scalable AI workloads on AWS SageMaker and Bedrock, utilizing Docker and Kubernetes for container orchestration and auto-scaling to handle peak traffic efficiently.
Thoughtworks
Sep 20202 yrs 4 mosDec 2022

Pune Division

Spearheaded the delivery of ASR systems for 18 Indic languages using PyTorch-based Wav2Vec 2.0, outperforming competitors by 15-20% WER through custom lexicon integration and data enrichment pipelines. Built TTS pipelines for 8 languages using GlowTTS and HiFi-GAN, achieving a 4.2 MOS score in production. Implemented BERT-based Inverse Text Normalization, improving transcription accuracy by 12% through robust post-processing techniques.
Thoughtworks
Feb 20207 mosAug 2020

Gurugram

GitHub
@ankurdhuriya

Ankur’s AI Stack

antigravity
antigravity
opencode
opencode
claudecode
claudecode

Education

Liverpool John Moores University

Master of Science - MS, Machine Learning & AI

Liverpool John Moores University

20232 yrs2025
DIT University

Bachelor of Technology - BTech, Computer Science

DIT University

20164 yrs2020

Join Ankur and 10k+ other professionals on

Tal Boss

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

Download Tal Boss

Ankur Dhuriya on Tal

Headline

AI Engineer @ ThoughtWorks | Customizing LLMs for Life Sciences & Procurement | Fine-tuning • RAG • RLHF | AI R&D | MLOps

About

Senior Applied AI Engineer with 6+ years of experience architecting enterprise-grade GenAI, RAG, and agentic applications. Expert in designing multi-step agent workflows using LangChain, LangGraph, and LlamaIndex, and implementing MCP-based context sharing for seamless enterprise interoperability. Proven track record in building robust ETL pipelines for document ingestion, developing secure microservices/APIs, and establishing comprehensive observability and evaluation frameworks. Proficient in Python best practices, CI/CD, and deploying scalable solutions on AWS Bedrock, Azure OpenAI, and open-source LLMs (LLaMA, Qwen).

Location

New Delhi

AI stack

antigravity

opencode

claudecode

Work experience

Senior Data Scientist at Thoughtworks, Jul 2025 - Present · 1 yr 2 mos, Gurugram, Architected and deployed a compliant, production-grade RAG chatbot for LifeScience Preclinical Safety team using AWS Bedrock, Claude Sonnet, and LangGraph; established end-to-end MLOps pipelines via GitHub Actions and Kubernetes with Weights & Biases for observability, implemented strict ALM protocols through SonarQube security gating, and developed a custom “Golden Dataset” evaluation framework that achieved 95% accuracy in a highly regulated pharmaceutical environment while streamlining access to critical research documentation. Agentic AI & MCP Integration: Architected a unified Copilot assistant using LangGraph and LlamaIndex, implementing MCP-based context sharing to enable seamless A2A integrations across Jira, Confluence, GitHub, and Databricks for real-time enterprise knowledge retrieval. Observability & Evaluation Frameworks: Established robust observability using LangSmith and DeepEval, implementing logging, tracing, and prompt tracking to monitor quality metrics. Achieved 94% accuracy in Biomedical QA via a self-reflective reasoning framework (Qwen3/LLaMA) with structured XML outputs. Performance Optimization: Designed a semantic caching layer with Redis, reducing latency from 6s to <800ms and cutting inference costs by 50% for 10K+ daily queries. Integrated Amazon Bedrock and Azure OpenAI to optimize model selection for cost/performance trade-offs. API & Microservices Development: Developed secure, scalable FastAPI microservices for AI inference, adhering to high-quality Python coding standards (modularity, testing, documentation) and ensuring seamless integration with frontend interfaces.

Senior Data Scientist at GEP Worldwide, Dec 2024 - Jul 2025 · 8 mos, Hyderabad, ETL & Document Pipelines: Built end-to-end ETL pipelines for document ingestion, parsing (OCR/IDP), and enrichment, processing 50K+ invoices/month. Implemented hybrid retrieval (BM25 + embeddings) over SQL/NoSQL databases and Vector Stores (Pinecone) for accurate RAG-based insights. RAG & LLM Integration: Engineered a spend analytics NLQ chatbot using LangChain and OpenAI/Claude models, achieving 92% answer relevance. Collaborated with SMEs to translate business needs into technical solutions, reducing AP cycle time by 35%. CI/CD & Productionization: Implemented robust CI/CD pipelines using Azure DevOps for automated testing and deployment of AI models to Azure Kubernetes Service (AKS), ensuring reliable production releases and version control for prompts and code. Security & Governance: Integrated custom guardrails for PII detection and prompt injection filtering, ensuring secure and compliant AI interactions in enterprise environments.

Data Scientist at Builder.ai, Dec 2022 - Dec 2024 · 2 yrs 1 mo, Gurugram, time by 70% (45min → 12min) with 94% syntactic correctness. Microservices & API Design: Designed and maintained robust FastAPI microservices for mul- tilingual transcription summarization, processing 10K+ calls/day across 8 languages. Ensured high-quality Python code with comprehensive unit tests and modular design. Observability & Monitoring: Implemented comprehensive logging and tracing for agent runs, enabling rapid debugging and performance tuning. Automated evaluation pipelines tracked halluci- nation rates and safety scores (99.2%), ensuring system reliability. Cloud Deployment: Deployed scalable AI workloads on AWS SageMaker and Bedrock, utilizing Docker and Kubernetes for container orchestration and auto-scaling to handle peak traffic efficiently.

Data Scientist at Thoughtworks, Sep 2020 - Dec 2022 · 2 yrs 4 mos, Pune Division, Spearheaded the delivery of ASR systems for 18 Indic languages using PyTorch-based Wav2Vec 2.0, outperforming competitors by 15-20% WER through custom lexicon integration and data enrichment pipelines. Built TTS pipelines for 8 languages using GlowTTS and HiFi-GAN, achieving a 4.2 MOS score in production. Implemented BERT-based Inverse Text Normalization, improving transcription accuracy by 12% through robust post-processing techniques.

Data Science Intern at Thoughtworks, Feb 2020 - Aug 2020 · 7 mos, Gurugram

Education

Master of Science - MS, Machine Learning & AI, Liverpool John Moores University

Bachelor of Technology - BTech, Computer Science, DIT University

Public links

github: https://github.com/ankurdhuriya