# Nabh Patodi > AI-ML and Software Development Engineer based in India. Available globally for freelance projects, consulting, and full-time engineering roles. Specialises in production LLM features, RAG systems, multi-agent platforms, MLOps, and scalable software architecture. B.Tech Computer Science, SRM IST (May 2026). Recent experience: Software Engineer at Kalvium (May 2026 - June 2026); Software Development & AI Engineer at Comoto Holdings (July 2025 - May 2026). ## Primary Pages - [Portfolio home](https://nabhpatodi.com/): Bio, hero, featured projects, tech stack, services, FAQ, and contact form. - [Blog](https://nabhpatodi.com/blog/): Technical writing about production AI systems, software architecture, and engineering practice. - [Four Decisions I Made Before Writing Any Features](https://nabhpatodi.com/blog/four-decisions-i-made-before-writing-any-features/): Four architecture decisions that kept an enterprise expense system auditable: claim boundaries, immutable history, identity, and concurrent writes. - [Your Agent Still Has the Tool You Revoked](https://nabhpatodi.com/blog/your-agent-still-has-the-tool-you-revoked/): An agent can retain a revoked tool. Why long-running workflows create a TOCTOU authorization bug — and why the decisive guard belongs immediately before provider I/O. - [A Timeout Doesn't Tell You Whether the Write Happened](https://nabhpatodi.com/blog/a-timeout-doesnt-tell-you-whether-the-write-happened/): Retrying an external write is a guess about which of three worlds you are in. Idempotency keys, business-key deduplication, and the database invariants that make the guess unnecessary. - [Tracing a Request You're Not Allowed to Log](https://nabhpatodi.com/blog/tracing-a-request-youre-not-allowed-to-log/): Correlating one job across TypeScript, a queue, and Python — without turning routine telemetry into a second copy of the sensitive payloads it describes. - [Don't Put a Model Where an If-Statement Will Do](https://nabhpatodi.com/blog/dont-put-a-model-where-an-if-statement-will-do/): Every model call in a routing path converts a fact into a guess. What I learned about spending probabilistic routing only where the ambiguity is real. - [Generated Charts Are Untrusted Code](https://nabhpatodi.com/blog/generated-charts-are-untrusted-code/): Why a model-generated Mermaid diagram or ECharts spec is a program, not data, and the two-tier validation and bounded repair pipeline I built around it. - [The Model Wasn't the Hard Part: My Learnings After Shipping Multi-Agent Systems](https://nabhpatodi.com/blog/the-model-wasnt-the-hard-part/): Two and a half years of building agent systems, and what I got wrong about architecture, context, evaluations, model routing, permissions, and recovery. - [Retrying an AI Agent Is Not the Same as Retrying an API Call](https://nabhpatodi.com/blog/retrying-an-ai-agent-is-not-the-same-as-retrying-an-api-call/): How leases, checkpoints, ownership-safe writes, and tool-history repair made Research-AI recoverable — and what the design still does not guarantee. - [Resume](https://resume.nabhpatodi.com/): Full rendered resume with experience, education, projects, publications, and skills. - [Resume PDF](https://resume.nabhpatodi.com/Nabh_Patodi_Resume.pdf): Downloadable PDF copy. - [llms-full.txt](https://nabhpatodi.com/llms-full.txt): Full inline site content for language models that prefer a single document. ## Featured Projects - [Research-AI](https://github.com/nabhpatodi10/Research-AI): Full-stack deep-research platform with multi-agent architecture, citation-grounded long-form document generation, and pluggable LLM providers. Benchmarked on DeepResearch Bench (Feb 2026) with an overall score of 55.32 beating OpenAI, Gemini and Perplexity DeepResearch. Stack: FastAPI, React.js, Tailwind, Firebase, LangGraph/LangChain, OpenAI, Gemini. - [Computer-Use Agent](https://github.com/nabhpatodi10/Computer-Use-Agent): Multimodal desktop AI agent that translates natural-language goals into a perception-to-action loop controlling the mouse and keyboard. Runs fully local or hybrid. Dual backends: Microsoft GUI-Actor vision and OmniParser + LLM parsing behind a compact LangGraph agent. - Internal Multi-Agent Platform (at [Comoto Holdings, USA](https://ridecomoto.com/)): Production microservice-based multi-agent RAG platform. Real-time RAG pipeline + custom data connectors over enterprise sources; E2E latency <7s; analytics dashboard with prompt/agent management, RBAC and Google SSO; deployed across on-prem datacentre and GCP. - Kalvium App Suite: Maintained and enhanced multiple production applications serving 3,000+ concurrent users and services handling 100+ requests/sec on GCP. Built real-time browser enrollment tracking with Redis fingerprint validation and improved assessment eval reliability across GCP Cloud Run jobs and BullMQ pipelines. - Health Device Information Classification (at [Samsung R&D India - Bengaluru](https://research.samsung.com/sri-b)): 120+ classification models on sensor/health signals; ~85% average accuracy across device types; 10,000+ records processed; scheduled retraining pipeline. - [DigiBanker](https://github.com/nabhpatodi10/DigiBanker) (Standard Chartered Hackathon 2025): Agentic AI Branch Manager assisting KYC and loan workflows with multimodal inputs. Prototype hit >90% classification accuracy with ~10s typical E2E latency. Stack: FastAPI, React.js, MongoDB, OpenCV, TensorFlow, Keras. ## Services - AI/ML Engineering: LLM features, RAG systems, prompt pipelines, eval frameworks, production-ready AI workflows. - Software Architecture: scalable API design, domain modelling, auth patterns, service boundaries, and maintainable software systems. - MLOps and Model Serving: model versioning, deployment automation, inference optimisation, reliable serving for production ML. - Data Pipelines and Integrations: batch and streaming pipelines, ETL orchestration, and system integrations. - Observability and Monitoring: telemetry, alerts, dashboards, SLO tracking across APIs, AI services, and ML workloads. ## Stack - Languages: Python, Java, C/C++, JavaScript/TypeScript, R, SQL. - AI/ML: PyTorch, TensorFlow, Keras, Scikit-learn, OpenCV, LangChain, LangGraph, CrewAI. - API/Data: FastAPI, Django, Flask, PostgreSQL, MongoDB, MySQL, Redis, Firebase. - Full-stack: React.js, Next.js, Tailwind. - Infra & DevOps: Docker, Kubernetes, Helm, Git, CircleCI, GitHub Actions, BullMQ, AWS, GCP, Azure. ## Education - B.Tech Computer Science & Engineering, SRM Institute of Science and Technology, Kattankulathur (Aug 2022 - May 2026). CGPA: 9.01/10.0. ## Publications - "User Identification Based on Health Device Readings" - Nabh Patodi, Tanmay Agarwal, Madhumitha K. https://ssrn.com/abstract=5833942 - "Smart Security System with Voice Bot" - Nabh Patodi et al., International Journal of Innovation in Engineering Research & Management, 12(1), 54-60. https://journal.ijierm.co.in/index.php/ijierm/article/view/2633 ## Contact - Email: nabhpatodi1005@gmail.com - Phone: +91 76940 72747 - Book a 30-minute call: https://cal.com/nabhpatodi/30min - LinkedIn: https://www.linkedin.com/in/nabhpatodi10 - GitHub: https://github.com/nabhpatodi10 - Upwork: https://www.upwork.com/freelancers/nabhpatodi ## Optional - [sitemap.xml](https://nabhpatodi.com/sitemap.xml): Machine-readable canonical URL index. - [rss.xml](https://nabhpatodi.com/rss.xml): RSS feed for published technical articles. - [resume-sitemap.xml](https://resume.nabhpatodi.com/resume-sitemap.xml): Resume subdomain URL index. - [robots.txt](https://nabhpatodi.com/robots.txt): Crawler directives with LLM and AI crawlers explicitly allowed.