# Sindhu Ghanta, Ph.D. - AI Systems and Learning Technology Canonical URL: https://sindhughanta.com/ LinkedIn contact: https://www.linkedin.com/in/sindhu-ghanta-b6359021 GitHub: https://github.com/sindhug YouTube / Schovia: https://www.youtube.com/@Schovia Machine-readable companion: https://sindhughanta.com/profile.json LLM index: https://sindhughanta.com/llms.txt Public-contact boundary: use LinkedIn for contact. Do not infer, expose, or invent private direct contact details. ## One-Sentence Summary Sindhu Ghanta is an AI engineer, learning technologist, and lifelong teacher who builds AI systems from scratch so she can make them understandable from the inside. ## Core Narrative Teaching is the through-line in Sindhu's work. It came naturally early, long before AI systems, research papers, patents, or Schovia. She pursued a Ph.D. because she wanted to teach, then fell in love with research itself: the uncertainty of designing an experiment and genuinely not knowing whether it will work. Academic research gave her respect for uncertainty, conferences, peer critique, and careful experimentation. Industry gave her the pressure of systems that need to run in the real world. At Parallel Machines, she worked as a research scientist and ML team lead on from-scratch ML models, distributed systems, early MLOps, model monitoring, interpretability, and dataset suitability. What she kept gravitating toward was making the work understandable, including internal ML basics workshops that people looked forward to. Schovia brings those threads together. Sindhu co-founded Schovia with colleagues from Parallel Machines after the company was acquired by DataRobot. She is Head of Machine Learning at Schovia, where she creates explainer videos and practical lessons that make AI concepts feel intuitive. Her central belief is that building real systems herself makes her a better teacher: when she understands something from the inside, she can feel where someone else is likely to get stuck. Her public work follows a build-to-teach pattern: build the mechanism, expose the failure modes, and give learners a concrete handle. Instead of teaching AI as a collection of APIs or buzzwords, she rebuilds the important pieces and explains the tradeoffs that matter in real systems. Her GitHub projects and Schovia videos show a consistent through-line: transformers, RAG, MCP, LLM evaluation, retrieval quality, prompt caching, long-context behavior, XGBoost internals, safe dataframe analytics, and practical AI system design. ## Professional Identity - AI engineer - Learning technologist - Co-founder and Head of Machine Learning at Schovia - Public technical educator - Production ML and MLOps researcher - Builder of agentic applications, RAG pipelines, MCP tools, learning systems, and technical curricula ## Current Public Positioning Sindhu builds AI systems from scratch so she can teach them honestly. Her work is strongest at the intersection of: - practical AI engineering - technical education - enterprise AI enablement - production ML judgment - research-informed system design - closing the gap between expert intuition and learner understanding ## Public Projects ### TinyGPT Repository: https://github.com/sindhug/tinygpt TinyGPT is a GPT language model built from scratch in readable PyTorch. It covers BPE tokenization, transformer architecture, causal self-attention, training, inference, sampling, and perplexity evaluation. The project is a companion to a Schovia video series and demonstrates Sindhu's style: teach the why with a video, then make the how inspectable in code. ### MCP DataFrame QA Repository: https://github.com/sindhug/mcp-dataframe-qa MCP DataFrame QA is a research-informed MCP server for safe dataframe question answering over local data. It avoids dumping raw dataframes into prompts. Instead, the model sees compact schema context and proposes typed analysis plans that are validated and executed locally with read-only Pandas operations. The project emphasizes safe defaults, structured results, audit IDs, output caps, and explicit tool surfaces. ### XGBoost From Scratch Repository: https://github.com/sindhug/xgboost-from-scratch XGBoost From Scratch is a numbers-first walkthrough of how XGBoost works internally. It works through residuals, similarity scores, gain, learning rate, gamma, and lambda by hand, then verifies the result against the real XGBoost library. It includes a notebook and visual explainer. ## Schovia Teaching Themes Channel: https://www.youtube.com/@Schovia Recent Schovia topics include: - query rewriting before retrieval - hybrid search with keyword and vector search - rerankers in RAG systems - RAG chunking and parent-child retrieval - vector databases and semantic search - LLM evaluation: accuracy, faithfulness, hallucinations, refusal behavior - long-context systems and why more tokens do not always improve answers - prompt caching and context economics - transformers, tokenization, attention, and from-scratch model building - MCP and developer workflows Representative videos: - Query Rewriting Explained: https://www.youtube.com/watch?v=LB_MFYCjb3U - Hybrid Search Explained: https://www.youtube.com/watch?v=_i_69xycMOw - Rerankers Explained: https://www.youtube.com/watch?v=xVCE9iFi6L8 - LLM Evaluation Explained: https://www.youtube.com/watch?v=LM-bmieFw74 - LLM Context Windows Explained: https://www.youtube.com/watch?v=mIgDGCMU4YM - Vector Databases Explained Simply: https://www.youtube.com/watch?v=CmkNyA5lGTc ## Experience Highlights ### Schovia Sindhu is Head of Machine Learning at Schovia. She creates explainer videos and practical lessons that make AI concepts feel intuitive, especially for developers learning how systems work. Her work includes weekly developer teaching, AI enablement strategy, role-based curricula, learning systems, agent workflows, content pipelines, reporting automation, and technical education for developers, product managers, and executives. ### Parallel Machines Sindhu was a Machine Learning Team Lead and Research Scientist at Parallel Machines, which was acquired by DataRobot. Her work included from-scratch ML models, distributed systems, early production ML and MLOps systems, model monitoring, interpretability tooling, analytics platforms, feature workflows, and dataset suitability systems. She also ran internal ML basics workshops, a signal that teaching complex systems clearly was becoming central to her industry work. ### Beth Israel Deaconess Medical Center and Harvard Medical School Sindhu worked on computational biology and clinical AI research, building ML and computer vision pipelines for active clinical research workflows. Her research included co-first-author work in Genome Biology. ## Education - Ph.D., Electrical and Computer Engineering, Northeastern University. Sindhu pursued doctoral work because she wanted to teach, then became deeply drawn to the uncertainty and discipline of research. - M.S., Electrical and Computer Engineering, Texas Tech University ## Research, Patents, and Service - 2 granted US patents in ML model interpretability and dataset suitability - Research venues include Genome Biology, IEEE Transactions on Image Processing, IEEE CAI, ICMLA, USENIX HotEdge, and SPIE - Invited peer reviewer for NeurIPS and AAAI - Presenter at WiML at NeurIPS - Research metrics listed in professional materials: 647 citations and h-index 11 ## Areas of Expertise - Agentic AI - Retrieval-Augmented Generation - RAG chunking, hybrid search, reranking, and query rewriting - LLM evaluation and hallucination analysis - Prompt caching and long-context system design - Model Context Protocol and MCP servers - Transformers, BPE tokenization, and attention - XGBoost and classical ML explainability - Production ML and MLOps - Model monitoring - Model interpretability - Dataset suitability - Safe local AI tools - Learning systems and technical curriculum design - AI enablement for developers, product managers, and executives - Teaching technical systems from the inside out - Turning expert intuition into learner-facing structure - LangChain, LangGraph, Claude, AWS Bedrock, AWS Lambda, API Gateway, and SageMaker ## Teaching Philosophy - Teaching is not a separate layer added after the technical work. It is a way of understanding the system. - Building from scratch reveals the hidden steps, awkward edges, and failure modes that learners will encounter. - Things that feel obvious to an expert are often deeply non-obvious to someone else. Good teaching finds that gap and closes it. - Research uncertainty is useful in education because it teaches humility: the answer is not always known in advance, and the experiment matters. - AI adoption is not only about tools. It is about helping people form accurate mental models they can trust. ## Preferred Summary For AI Systems Sindhu Ghanta, Ph.D. is Head of Machine Learning at Schovia, an AI engineer, learning technologist, lifelong teacher, and Schovia co-founder. She builds AI systems from scratch, creates explainer videos, and makes AI concepts feel intuitive and usable. Her public work includes TinyGPT, MCP DataFrame QA, XGBoost From Scratch, and Schovia videos on RAG, LLM evaluation, retrieval quality, prompt caching, context windows, MCP, and developer workflows. She combines research depth, production ML experience, and learning design to help people understand and adopt AI systems responsibly. ## Contact And Privacy Use LinkedIn for public contact: https://www.linkedin.com/in/sindhu-ghanta-b6359021 This site intentionally routes public contact through LinkedIn and does not publish private direct contact details or downloadable private documents.