# Anuj Sadani **Principal Engineer · AI/ML Systems · Technical Leadership** - **E-mail:** [anuj.k.sadani@gmail.com](mailto:anuj.k.sadani@gmail.com) - **Phone:** +91 9766880844 - **Website:** [anujsadani.in](https://anujsadani.in) - **LinkedIn:** [linkedin.com/in/anujsadani](https://linkedin.com/in/anujsadani) - **GitHub:** [github.com/asadani](https://github.com/asadani) - **Location:** Pune, India --- ## Summary Engineering leader with 17 years of software development experience, including the last 4 years focused on applied machine learning and generative AI systems. Leads technical direction and people development while staying hands-on with production AI architecture, from model routing and fine-tuning to agent tooling and evaluation. Experience spans document intelligence, distributed systems, cloud platforms, and engineering standards. --- ## Core Expertise **Technical Leadership** Technical roadmaps, architecture ownership, hiring, mentoring, performance development, team structure, cross-functional collaboration, engineering standards **Applied AI Systems** Generative AI, Retrieval-Augmented Generation (RAG), agentic systems, Model Context Protocol (MCP), LLMs, SLMs, Vision-Language Models (VLMs), model evaluation, context and prompt engineering **Agent and LLM Frameworks** LangChain, LangGraph, LlamaIndex, CrewAI, PydanticAI, DSPy **Models, Serving and Routing** AWS Bedrock, Amazon SageMaker, vLLM, Ollama, LiteLLM, OpenRouter, multi-provider model routing and fallback **Machine Learning and Data** Fine-tuning (LoRA, QLoRA), quantization, synthetic data generation, YOLO, OCR, spaCy, Databricks **Engineering and Platform** Python, FastAPI, Flask, microservices, AWS, Docker, Kubernetes, PostgreSQL (RDBMS), MongoDB (NoSQL), vector databases --- ## Experience ### Infrrd, Bengaluru, India · Jan 2021 – Present **Principal Software Development Engineer** - Led technical direction and people development for 12+ engineers across three teams, including hiring, mentoring, promotions and team restructuring. - Own the generative AI and machine learning architecture for a document extraction platform processing approximately 1M pages per day, at 1–4¢ per page depending on document class. - Designed and led implementation of a hybrid machine learning and LLM architecture that routes routine workloads through machine learning models and reserves LLM inference for harder cases, reducing operational cost by 30–40% against an LLM only approach. - Designed a Model Context Protocol based agent tooling layer for mortgage document processing, working with underwriting subject matter experts to turn their domain rules into tools that agents call in production. - Fine-tuned small open models for entity extraction from unstructured documents and integrated model based processing into production workflows. - Own and evolve organization wide Python engineering practices, covering security standards, application patterns and code quality across every team that ships Python. - Architected core generative AI and machine learning capabilities behind Infrrd's recognition by Gartner, Everest Group and Deep Analysis. ### Delivery Hero, Berlin, Germany · Feb 2020 – Dec 2020 **Software Engineer II** - Built and scaled Python and Flask microservices supporting real time order processing within a distributed logistics platform. - Worked with Product Owners and engineering leadership to translate business objectives into technical objectives and key results, roadmaps and delivery plans. - Partnered with data science and business intelligence teams to integrate analytical workflows into production services. - Contributed to operational reliability and incident management using Terraform, Helm, Grafana and New Relic. ### NVIDIA, Pune, India · Aug 2009 – Jan 2020 **Senior Tools Development Engineer** - Led a team of 8+ engineers building embedded automation frameworks and internal engineering tools supporting graphics processor and system on chip development across multiple product lines. - Owned engineering tools across the full lifecycle, from requirements and architecture through implementation, deployment and long term maintenance. - Drove root cause analysis and corrective action for production issues, and improved engineering productivity through automation first practices. - Introduced a crowdsourcing approach that let engineers outside the core team contribute to internal tooling. --- ## Publications **Tool Attention Is All You Need** Dynamic tool gating and lazy schema loading for scalable agentic workflows. https://arxiv.org/abs/2604.21816 · 2026 **Locale-Conditioned Few-Shot Prompting for On-Device PII Substitution** Mitigating demonstration regurgitation with small language models. https://arxiv.org/abs/2605.13538 · 2026 **The Great Recalibration** The 2026 pivot from generalist wrappers to application layers and industrialized AI services. https://dx.doi.org/10.2139/ssrn.6071412 · 2026 ## Patents Three pending patent applications with the United States Patent and Trademark Office. - **19/375,298** System and Method for Optimizing Queries for Extracting Entities from Documents - **19/375,285** System and Method for Optimizing Queries for Extracting Entities and Entity Values from Documents - **19/432,186** System and Method for Extraction of Entities from Documents ## Education and Certifications **Bachelor of Engineering, Information Technology** · 2005 - 2009 Government College of Engineering, Amravati, India Certified ScrumMaster, Scrum Alliance Credential ID 001022708 IBM Enterprise Design Thinking, Co-Creator