Aniket Kadukar

Serving Notice Period l AI Engineer with 6+ years of experience building AI-powered systems, from backend APIs to production LLM pipelines.

Experienced across the full lifecycle: requirements scoping, data pipeline design, model fine-tuning, and production operations. Recent focus on RAG systems, LLM evaluation frameworks, and agentic workflows, with deployments on Azure Databricks and Kubernetes. Focused on building systems that are reliable and cost-efficient in production, not just impressive on paper.

Senior Consultant at Capgemini (2025-07 – Present)

• Built an MCP server for LLM-based evaluation workflows and developed the client-side APIs to consume it, handling the full request-response cycle end to end.
• Deployed GenAI services on Azure Databricks, containerizing workloads and configuring GPU clusters to keep inference costs predictable.
• Implemented authentication and role-based access control using Azure Entra ID (OAuth 2.0) across all service endpoints, meeting enterprise governance standards.
• Built a distributed web scraping and RAG pipeline that ingested 2 lakhs+ web pages; parallelization made harvesting 300% faster, and agentic semantic chunking with metadata injection improved retrieval precision by 20%.
• Set up end-to-end observability with LangSmith and MLflow on Azure Databricks, tracking prompt versions, latency, and output quality across three production systems.
• Reduced LLM inference costs by 65% and improved throughput 4x by replacing a 70B model with a distilled 8B model using INT4 quantization (bitsandbytes/AWQ), while keeping output quality comparable.
• Built an LLM-as-a-Judge evaluation system on the evaluation platform that replaced 92% of manual QA work, reaching a 0.88 Cohen's Kappa alignment score against human reviewers using few-shot prompting.
• Wrote application-layer LLM routing logic that dynamically selected models based on query complexity, cutting API costs by a further 35%.
• Reduced end-to-end RAG latency by 40% through tracing and bottleneck analysis across the retrieval layer.
• Managed infrastructure across Azure Databricks and Kubernetes with Conductor for workflow orchestration, maintaining 99.9% uptime across deployments.

Associate at Cognizant (2020-01 – 2025-07)

• Built AI-powered backend systems integrating LLM capabilities with production APIs, starting with FastAPI and Django and evolving into multi-agent patterns.
• Designed multi-agent workflows using LangGraph for automation tasks requiring human-in-the-loop checkpoints and conditional branching.
• Developed IoT integration solutions across AWS IoT Core, GCP IoT Core, and Azure, handling device authentication with JWT and x.509 certificates.
• Built a FastAPI service for automated bug reporting that used an LLM to convert raw session data into structured bug reports via prompt chaining and few-shot examples.
• Used Pydantic with LLMs to enforce structured output generation, significantly reducing downstream parsing errors.
• Integrated LangChain for LLM interfacing with close attention to token usage and prompt efficiency, keeping costs manageable at scale.
• Implemented a Neo4j knowledge base to power a graph-aware GenAI application, supporting relationship traversal in retrieval.
• Built a YOLOv8 computer vision pipeline for real-time AR object tracking, serving results via Django REST APIs with async video processing through Celery.
• Developed backend APIs for a 6-axis robotic arm, training ML models to interpret natural language commands and map them to executable robotic actions.
• Containerized applications using Docker with full CI/CD pipelines via GitLab CI, from development to production.

B.E. in Electronics and Telecommunication Engineering – Shri Sant Gajanan Maharaj College of Engineering, Shegaon, MH (2015-08 – 2019-05)