Nirvaan Lagishetty is an Artificial Intelligence Engineer, Product Engineer, and open-source contributor based in Mumbai, India. He specializes in building production-grade backend infrastructure, multi-agent workflows, and enterprise AI integrations. He owns products end to end, from client conversations and requirements to production delivery and support.
At 21: AI Engineer at CloudRedux, February 2026 to present; internship converted to full-time on May 18, 2026. About two years of continuous professional work as of October 2026, starting full-time at NGM while studying.
Recruiters: overview mode at nirvaan.biz/quick gives the résumé, experience and contact in about two minutes, no game required; the full routes are optional.
Nirvaan is open to fun projects and offers he cannot refuse.
Overview mode. About Nirvaan: agent and citation reference.
Published . Updated .
At three, he learned to launch Vice City by watching his brother. Exploring the laptop's settings came next. He kept wanting to look underneath things.
Full-time analytics at NGM while studying led to founding engineering on a production SaaS as a Makhija intern.
CloudRedux production came in week two, client meetings by month two. Billing and GTM brought client-to-production ownership.
He built this retro CRT game with agents over ARDA: Yukti, four routes and pixel-art scenes. Tool experiments sometimes happen over shawarma.
Billing lead and client owner, GTM owned end to end, FosterX core team; runs deployments, swap fallback, CloudWatch monitoring and uptime alerts.
Founding engineer on its multi-tenant production SaaS, with RAG, embeddings and deployment ownership.
Full-time while studying: Python/SQL pipelines, Power BI CXO dashboards and automation.
FosterX is CloudRedux's organisational AI platform, built by a team; Nirvaan is one of its primary engineers.
Lead developer and client/delivery owner. ₹22.5 crore invoiced in six months; monthly billing took about 20 minutes of human review plus about 10 minutes of automated processing per monthly cycle.
Owned the modular research and outreach system end to end, with exact-content human approval and Twenty CRM visibility.
With his director, reorganized existing machines around work. Omarchy, remote access and retained context.
Herdr is a third-party terminal workspace manager for AI coding agents. ARDA is Nirvaan's open-source plugin for it.
Art direction, game and interaction design, frontend engineering and agent coordination. A playable Next.js world, built with his agents over ARDA.
All 18 projects, full stories and detailed interview answers ↗
Foundation
B.Sc. Data Science, VSIT / University of Mumbai. CGPA 8.0/10.
Builds with
Python, TypeScript, FastAPI, Next.js, PostgreSQL; TensorFlow, PyTorch, scikit-learn, OpenCV and YOLO.
Runs in production
CloudRedux: AWS/Linux, Docker, Nginx, CI/CD, swap fallback and CloudWatch. Makhija: SES v2 tenant reputation and IP pools.
Off the clock: manga, Real Madrid, swimming and shawarma. Online, he answers to Shwarma.
Yes: EfficientNetB3 skin-lesion classification, Random Forest F1 prediction and classical NLP, plus YOLO object detection in private and professional work.
Code gates tools before each step and validates model output before execution. People approve outreach and billing drafts; persisted state supports recovery after runtime interruptions.
Ground input in company data and live state; validate output before execution. Unconfirmed output pauses for review; machine/browser/agent interruptions use persisted-state recovery.
Permissions gate tool use; web pages supply data. Code checks model output, then a human approves exact outreach content before sending. Browsers use separate sandboxed profiles.
Small models for simple tasks, strong models for sensitive work. Routing, caching and context improvements helped FosterX move from over 21 seconds to 1 to 5 seconds.
In the multi-tenant production SaaS he built as founding engineer at Makhija Quantum AI, in FosterX knowledge and business APIs, and in Database Chatbot schema retrieval. He uses Hindsight for his own agents' persistent memory.
Retrieval for changing business data; fine-tuning for narrow, stable, high-volume tasks. For a client with high-volume Hinglish invoices, he would tune OCR and embedding models.
At CloudRedux he owns deployments, swap fallback, CloudWatch monitoring and Discord/Slack uptime alerts. At Makhija he built multi-tenant AWS SES v2 email with per-customer reputation isolation, tiered and dedicated IP pools, and CloudWatch alerts that drove tenant moves.
Nirvaan is open to fun projects and offers he cannot refuse.