# Anu Kumar — Senior AI Engineer > I build agent systems that survive production. Senior AI Engineer working on LLM agents, retrieval pipelines, and the infrastructure that keeps them running. **Location:** Mohali, Punjab, India **Email:** anu2614kumar@gmail.com **GitHub:** https://github.com/trickster26 **LinkedIn:** https://www.linkedin.com/in/anukumar0 ## What I do ### Agent architecture Reasoning loops, tool-calling, confidence scoring, and human-approval checkpoints. The interesting problems are not the prompts. They are what happens when a tool call fails at three in the morning, when a model returns something plausible and wrong, and when a decision needs a person to sign off before it executes. - Multi-step tool-calling agents - Confidence scoring on agent output - Human-in-the-loop approval gates ### Retrieval that grounds RAG pipelines that tie agent decisions to real company data instead of model memory. Chunking, embedding, ranking, and the evaluation loop that tells you whether retrieval is actually helping. ### The part that stays up FastAPI and Node services, PostgreSQL and Redis, Docker and Kubernetes, CI/CD. Models are a component. The system around them is the product. ## Selected work ### SuperManager (2026) The agent and LLM layer of an enterprise agentic platform. Core contributor to the agent runtime: reasoning and tool-calling workflows, confidence scoring on generated output, and human-approval checkpoints before an agent action commits. The retrieval side grounds those decisions in company data rather than model priors. **Stack:** Python, FastAPI, LLM agents, RAG, Docker, Kubernetes ### NapCRM (2025) Multi-tenant SaaS CRM where every customer gets an isolated database and its own process. Provisioning creates the tenant database, runs industry-specific migrations, starts a dedicated backend process, and writes the DNS record. Feature access is gated per industry and plan, so one codebase serves verticals that do not resemble each other. **Stack:** Node.js, Express, MariaDB, React, PM2, Cloudflare ### Adnosaur (2024) Product and ad-campaign discovery for dropshippers, with an AI-assisted store search. Full-stack build on React and Node with a FastAPI service alongside it. Integrated the OpenAI realtime API so users could interrogate the catalogue conversationally instead of filtering it. **Stack:** React, Node.js, PostgreSQL, FastAPI, AWS, OpenAI ### NapSEO (2025) An SEO command centre run by a set of specialised agents rather than one prompt. Separate agents handle audit, keyword research, content briefs, and internal linking, each with its own tools and output contract. Multi-project, so one workspace can hold unrelated sites without the agents bleeding context across them. **Stack:** Node.js, React, LLM agents, PostgreSQL ### Covix (2022) A convolutional model that classifies chest X-rays for COVID-19 positivity. Trained end to end: dataset preparation, augmentation, architecture selection, and evaluation. The early project that made the case for everything after it. **Stack:** Python, CNN, Computer vision, Deep learning ## Multi-tenant architecture ### One codebase. Isolated tenants. No shared query path. Multi-tenancy usually means a tenant_id column and a lot of discipline. NapCRM does not work that way. Each customer gets a physically separate database and its own running process, which makes a cross-tenant leak something you cannot write by accident. - **Provision:** A dedicated database is created and migrated with the schema for that customer's industry. - **Isolate:** The backend boots bound to one database pool. No code path reaches another tenant's data. - **Route:** A process starts under its own supervisor entry, and a DNS record points the customer's subdomain at it. - **Gate:** Industry and plan flags decide which modules load, so one deployment serves verticals with nothing in common. ## Experience ### Senior AI Engineer, SuperManager AGI _Mar 2026 - Present (current)_ Designing agent and LLM systems for project intelligence, workflow automation, and decision support. Building RAG pipelines with confidence scoring and human-approval controls, wired into enterprise tools and databases, shipped through Docker, Kubernetes, and CI/CD. **Stack:** Python, FastAPI, LLM agents, RAG, Kubernetes ### Full Stack Developer, Mind2Web _Aug 2023 - Feb 2026_ Built full-stack applications on React, Node.js, and PostgreSQL across DigitalOcean and AWS, with REST APIs for real-time data exchange. Engineered an AI-assisted interaction platform on React, FastAPI, the Twilio SDK, and the OpenAI APIs. **Stack:** React, Node.js, PostgreSQL, FastAPI, Twilio, AWS ### Developer, remote, Zirizi.AI _2023_ Django and React applications with secure backends on the Django ORM, SQLite3, and AWS EC2. **Stack:** Django, React, AWS EC2 ### Developer, Chandigarh, QSpiders _2023_ Java applications on JDBC and J2EE, plus SQL query optimisation that cut turnaround on the team's reporting work. **Stack:** Java, JDBC, J2EE, SQL ## Stack - **AI and ML:** AI agents, LLM applications, RAG, NLP, Prompt engineering, OpenAI API, Computer vision, Deep learning - **Languages:** Python, JavaScript, TypeScript, Java, C++, PHP - **Backend:** FastAPI, Node.js, Express, Django, Flask, J2EE - **Data:** PostgreSQL, MySQL, MariaDB, MongoDB, Redis, Oracle - **Platform:** AWS, Docker, Kubernetes, CI/CD, Cloudflare, Git ## Education B.Tech, Computer Science and Engineering, I.K.G. Punjab Technical University, 2019 - 2023, Mohali, Punjab. ## Contact ### Working on something that needs an agent layer? Open to senior AI engineering roles, and to problems where the model is the easy part. The fastest way to reach me is email. Email: anu2614kumar@gmail.com --- More: https://anukumar.site/about.md · https://anukumar.site/contact.md · https://anukumar.site/privacy.md · https://anukumar.site/llms.txt