Engineering AIplatforms, developertools & the systemsthat ship them.
Agentic AI · MCP Servers · Kubernetes Infra
Prakhar Sharma-Full-Stack / AI Engineer @ Katonic AI
I design and ship production-grade web apps, AI platforms, and the developer tooling that runs them - turning complex systems into interfaces that feel effortless.
I turn complex systems into products that feel effortless.
Full-Stack / AI Engineer at Katonic AI, working where AI meets infrastructure - the platforms, agents, and developer tooling that keep AI systems dependable in production.
Day to day that spans the whole stack: React and Node product surfaces, MCP servers and agent tooling, and the Kubernetes infrastructure underneath - GPU allocation, multi-tenancy, and Prometheus-based monitoring. Outside work, I build full-stack products end-to-end, from developer tools to commerce.
Prakhar Sharma
Full-Stack / AI Engineer
Pune, Maharashtra, India
- Currently
- Engineer @ Katonic AI
- Building
- AI platforms & MCP tooling
- Open to
- Full-time & freelance
Engineering across the whole stack.
Not a list of logos - the value I create, and what I've built doing it.
Frontend Systems
Interfaces that make complex products feel effortless.
Backend APIs
Reliable APIs that scale with products, not prototypes.
AI Integrations
LLMs and agents wired into products people actually ship.
Developer Tooling
Tools that remove friction and make teams faster.
Infrastructure
Deploys that stay boring, repeatable, and safe.
Databases
Data models that hold up as products grow.
Currently deepening my AI Engineering stack
I already build on real AI platforms and production systems. Now I'm going deeper into the engineering foundations behind modern AI - Python and ML fundamentals, LLM application architecture, RAG, agents, and evaluation.
Shipping real software, in production.
Full-Stack / AI Engineer
Katonic AI · Remote
The challenge
Take an MLOps and AI platform from single-tenant to multi-tenant production - spanning the React/Node product surface, Kubernetes infrastructure and GPU scheduling, and a shift from generative to agentic AI.
What I owned
- Developed and operated Apps, Model Deployments, and Workspaces end-to-end across the full stack on Kubernetes-based infrastructure.
- Built a monitoring dashboard for resource usage, storage volumes, and network performance from Prometheus metrics for real-time operational visibility.
- Deployed MCP servers and integrated them with AI agents to enable tool calling and workflow automation.
- Built text-to-speech into the chat system to make conversational AI experiences feel more natural.
Engineering decisions
- Implemented GPU slicing and allocation across multi-tenant and single-tenant architectures to optimize workspace and model-deployment resource utilization.
- Implemented agent guardrails and policy controls to keep agentic workflows reliable, secure, and governable.
- Contributed to migrating the platform from single-tenant to multi-tenant to improve scalability and collaboration.
- Contributed to the platform's shift from generative AI workflows toward an agentic AI architecture.
Impact
- Gave teams real-time visibility into compute, storage, and network health.
- Improved platform scalability and collaboration through the multi-tenant migration and tuned GPU allocation.
- Raised platform stability by debugging issues across frontend, backend, APIs, deployments, MCP integrations, and infrastructure.
Lessons learned
- Multi-tenancy is an architecture decision, not a feature - isolation and resource limits have to be designed in from the start.
- Agent reliability comes from guardrails and observability, not from the model alone.
Technologies
Full Stack Engineer
InternshipKatonic AI · Remote
What I owned
- Developed and maintained full-stack applications built on generative-AI technologies.
- Designed and implemented responsive React and Next.js interfaces for AI-driven platforms, tuned for smooth interaction and performance.
- Built a file manager on the MERN stack with NFS and Kubernetes, enabling centralized, scalable dataset storage without external tools.
- Collaborated with cross-functional teams to design and implement scalable solutions.
Impact
- Optimized backend services for AI-powered platforms.
- Improved the end-user experience across AI-driven products.
Technologies
Education
B.Tech - Computer Science (AI & ML)
Oriental Institute of Science and Technology · Bhopal, India
2021 - 2025
The path here
- 2021
The first commit
Fell for the web - turning ideas into things people could actually click and use.
- 2022
Thinking in systems
Moved from pages to products - components, state, and APIs that fit together.
- 2023
Full-stack, end to end
Shipped complete MERN applications - auth, data, and polished frontends in production.
- 2024
Production engineering
Focused on scalable architecture, cloud deployments, and backend systems people actually relied on.
- 2025
AI platform engineering
Joined Katonic AI - building AI platform surfaces on Kubernetes-based, GPU-scheduled, multi-tenant MLOps infrastructure.
- 2026Current
Agentic AI & AI Engineering
Shipped MCP servers and agentic AI at Katonic, and started going deeper into the foundations behind intelligent systems - Python, LLM applications, RAG, evaluation, and model-driven workflows.
Currently building toward
Going deeper into AI Engineering - from Python and ML fundamentals to LLM systems, RAG, agent architectures, evaluation, and production AI.
Products I've designed and shipped.
A featured deep-dive, then a selection of things I've built end-to-end.
MERN-AI-GPT
A full-stack AI chat platform with authentication, streaming LLM responses, and persistent conversation history.
The problem
Most side-project chat apps feel like demos - no streaming, no history, no auth, and a UI that gets in the way.
Why I built it
I wanted to build the full stack of a real AI chat product: authentication, a streaming interface, and durable history - not just a prompt box wired to an API.
Architecture & data flow
Technical decisions
- Streamed responses over the wire so the interface feels alive instead of waiting on a spinner.
- Stored conversations in MongoDB keyed per user for durable, resumable history.
- Kept a clean React client / Node API boundary so each side could evolve independently.
- Added JWT authentication so it behaves like a real, multi-user product.
Engineering challenges
- Rendering streamed tokens smoothly without layout thrash.
- Keeping conversation state consistent between the client and stored history.
Production learnings
- Streaming is 20% of the work and 80% of the feel.
Results
- A complete full-stack AI app - auth, streaming, and persistence - rather than a demo.
- A chat experience that feels immediate and stays out of the way.
What's next
- Tool calling and retrieval over user documents
JobVault
A full-stack job-search workspace built around Kanban workflows, application tracking, resume management, and analytics.
CodeVitals
An AI-powered repository health toolkit that analyzes codebases and exposes actionable diagnostics through an npm package and MCP interface.
SeepCraft
A production-focused e-commerce storefront for handcrafted home decor, engineered around reusable commerce components and a polished shopping experience.
Tempest
An award-inspired cinematic web experience engineered around motion, visual hierarchy, and scroll-driven storytelling.
How I engineer, in the open.
Building through production software, developer tooling, and continuous experimentation - here's the activity behind it.
Contribution activity
143 in the last yearTop languages
- JavaScript48%
- TypeScript19%
- HTML14%
- Java9%
- CSS8%
Current focus
Pinned technologies
Open-source highlights
- Published npm package
- Built an MCP server
- Production AI platform
- Full-stack SaaS
- Freelance commerce platform
- Award-inspired UI
Certifications & recognition.
AI Platform Engineering
Katonic AI
Developer Tooling & MCP
Open source
Open-Source Contributions
GitHub
Google IT Support Certificate
Google · Coursera
NPTEL Certification
NPTEL
Hackathon Participant
Multiple events
Let's build something great.
Have an idea, a role, or a hard problem? I'd love to hear about it.