Headline

Senior AI Engineer @ Toast ✦ Previously @SAP Labs · @Accenture ✦ Building AI Systems that survive production, not just demos ✦ Generative AI · Agentic AI · Machine Learning · Artificial Intelligence

About

Most AI demos work. Most AI in production doesn't. I work on that gap. I'm a Senior AI Engineer on the Ads & Insights team at Toast, building agentic and LLM systems that have to survive real traffic — where "it worked in the notebook" is the start of the problem, not the end. Before Toast, I helped build SAP's Joule copilot, where I co-invented a patent-pending AST-based code-context retrieval framework: deterministic code understanding for AI agents that beat semantic search on completeness (95% vs 65%) while cutting token cost per call by ~60%. Earlier, at Accenture, I shipped RAG systems and GenAI tooling across enterprise clients. The foundation under all of it is classical ML — I was training and shipping models before "GenAI" was even a category. That's exactly why I can tell demo-ware from something that'll actually hold up in production. What I actually obsess over — the stuff that decides whether an agent is a product or a retry script: → Retrieval quality over bigger models → Eval harnesses and benchmarking → Token economics → State machines over "autonomous" loops I write about the real version of building with LLMs here — what works, what's hype, and what it actually costs. If that's your kind of thing, follow along. Stack I live in: Agentic AI · LLMOps · RAG · Python · AWS · Classical ML · evaluation & benchmarking. Building something in this space? Let's talk. Portfolio's in my Featured. 👇

Location

Bengaluru

AI stack

anthropic apis

claude code

deepseek

langfuse

langgraph

huggingface

Work experience

Senior AI Engineer at Toast, May 2026 - Present · 5 mos, Bengaluru, • Building AdTech Intelligence • Leading AI feature strategy for the Ads and Insights products.

AI Engineer II at SAP, Mar 2025 - May 2026 · 1 yr 3 mos, Bengaluru, Building SAP Joule • Creating SAP’s flagship agentic solution, working as a copilot embedded in the SAP product ecosystem, providing seamless connectivity and a unified experience across products. • Co-invented a patent-pending AST-based code context retrieval framework enabling AI agents to deterministically extract code context and dependencies, replacing grep-based pattern matching and semantic search. Showing 25% first-pass code accuracy lift, 60% token consumption reduction per agent call, 95% retrieval completeness vs. 65% semantic baseline. Preceded similar capabilities in Copilot Workspace by 6+ months. • Worked on Joule for Developer (J4D) intended for accelerating migration of legacy ECC code to the S/4 Hana cloud environment by implementing an end-to-end code generation pipeline, reducing migration timelines by 35% • Fine-tuned open-source foundation models for specific use cases, including text-to-code and code-to-code use cases. • Created a benchmarking framework for code quality assessment with dynamic unit tests, syntax checks and LLM-as-a-judge manual evaluation dependency, improving feature shipping time by 15%

Machine Learning Engineer at Accenture, Sep 2023 - Mar 2025 · 1 yr 7 mos, Bengaluru, • Developed a procedure to transform the existing content of an enterprise-level website into an indexed vector database, creating a knowledge base for a RAG-enabled chatbot that served as a quick search tool and cut down search time for users by 15%, significantly boosting user retention. • Designed and implemented a content approval system for a pharmaceutical client, enabling regulatory and legal approvers to evaluate and approve campaign content being published on the website. Reduced approval time per page from 4 hours to 2.5 hours, directly reducing go-live time and marketing campaign timelines. • Built a risk assessment model for predicting at-risk deliverables, increasing time-bound objective delivery by 25%, especially in remote working projects. • Collaborated with over 12 projects to create standardised runbooks for the induction of generative AI tools, significantly enhancing work efficiency while ensuring data integrity and compliance. • Deployed generative AI POCs in working with large language models such as Llama-2/3, Mistral, and GPT-3.5/4, augmenting with a vector database with frameworks like LangChain and Streamlit

Associate Software Engineer at Accenture, Jun 2021 - Sep 2023 · 2 yrs 4 mos, Bengaluru, • Built a content management system with multilingual support for no-code content authoring, targeting campaigns, and executing workflows with custom components. • Audited and refactored the existing application for bottleneck fixes that would yield maximum improvement, resulting in an increased page performance score (Lighthouse) by 135% with a 30% tech debt reduction. • Migrated rich media hosting Scene7 to enhance media processing capabilities, accelerate image loading, and remove dependency on device-specific image renditions, resulting in a 50% drop in the size of media-heavy pages • Developed a server cleanup utility for clearing servers of long-dormant pages and media to save space and keep servers lightweight, leading to a 15% server bloat reduction. • Engineered an ITSM management framework, revamping ITSM handling procedures across the project, resulting in a 60% drop in incident/service request response time and over a 90% drop in active incidents/tickets.

Machine Learning Instructor at Edurific, Jul 2020 - Mar 2021 · 9 mos, Mumbai, Maharashtra, India, ● Worked as an AI/ML Instructor and implemented end-to-end Machine Learning projects for integrating into courses. ● Libraries worked with: Scikit Learn, TensorFlow, NumPy & Pandas, Keras, Flask, Streamlit. ● Project Worked on: Time series analysis of Airways miles. Face mask detection module. ● Delivered sessions on concepts in ML application development along with programming requisites. ● Built Dialogflow Chatbot for streamlining content delivery. ● The projects created impacted over 700 students.

Education

Bachelor of Engineering - BE, Computer Engineering, K. J. Somaiya Institute of Technology

Junior College, Dadasaheb Tirodkar Jr. College

Public links

github: https://github.com/naikshubham14

shubhamnaik.vercel.app: https://shubhamnaik.vercel.app/