Ayush Kumar
Gupta.
B.Tech student in Electronics & Computer Science Engineering at Bhilai Institute of Technology, building backend systems, AI agents, and automation pipelines that hold up under real load — not just demo well.
System-minded engineer, agent builder
I work across the stack, but the part I enjoy most is the backend and the intelligence layered on top of it — RESTful APIs, database design, and LLM-powered automation that has to behave predictably even when the input doesn't.
Recent work spans MERN applications in production-style settings (an EV rental platform, full-stack client projects), and independent systems like a self-healing browser automation agent and a retrieval-augmented document assistant. I like projects where the AI has to make a decision and the system has to catch it if that decision is wrong.
Currently expected to graduate in May 2027, and picking up hackathon wins along the way.
Where I've shipped code
- Built AI-integrated MERN applications for a campus EV and two-wheeler rental platform.
- Designed RESTful APIs in Node.js and Express for authentication, booking, and vehicle management.
- Built responsive React interfaces and optimized MongoDB schemas for faster reads and writes.
- Integrated an AI chatbot into the support flow and collaborated via Git in an Agile team.
- Developed and deployed full-stack MERN applications from spec to production.
- Built secure REST APIs with Node.js/Express and responsive React interfaces.
- Designed MongoDB schemas focused on query performance and data integrity.
- Integrated an AI chatbot for user assistance and handled debugging and performance tuning.
Systems I've built end to end
OmniAgent
STATUS: ACTIVEA full-stack AI desktop automation agent that turns plain-language instructions into structured, multi-step execution plans — run through Groq, Claude, and GPT-4o, then carried out on screen.
Its browser engine escalates through five detection tiers, from DOM matching to accessibility-tree analysis to Gemini vision, so a broken selector doesn't break the task. Every generated plan passes through a risk-classification and safety review layer before it's allowed to execute.
AI RAG PDF
STATUS: SHIPPEDA document interaction platform that lets you ask questions of a PDF and get context-aware, citation-backed answers instead of a generic summary.
Handles upload, text extraction, embedding generation, and vector indexing, then routes retrieved passages through an LLM for grounded, source-linked responses.
Real-Time Voice Translator
STATUS: SHIPPEDA speech-to-speech translation system built for live conversation — spoken input goes in, translated speech comes out, with latency tight enough to hold a real exchange.
Optimized the audio streaming and inference pipeline and tuned the recognition model to keep accuracy high without giving up response time.
Tools of the trade
Languages
Frontend
Backend
Databases
Cloud & DevOps
Machine Learning
Tools
Hackathons & recognition
Hackathon Winner — Fintech Security Bootcamp
IIT Bhilai
Hackathon Runner-Up — National Hackathon
IIIT Naya Raipur
Hackathon Participant & Builder
GCE — shipped under time pressure alongside a team
Let's build something
Open to backend, AI/ML, and full-stack roles and collaborations. The fastest way to reach me is email — I check GitHub and LinkedIn too.