Headline

SDE Intern @ Xelron | AI Agents, RAG & LLM Evaluation | Full-Stack: Python, React, Node.js, Next.js | MNNIT Allahabad ’26 | Codeforces Specialist | Open to Full-Time SDE / AI Engineer Roles 2026

About

Software Engineer building AI agent systems, RAG/LLM applications, and full-stack products — currently an SDE Intern at Xelron, where I benchmark and evaluate LLM-based coding agents. I graduate from MNNIT Allahabad (B.Tech CSE, 2026) and I'm actively looking for full-time Software Engineer / AI Engineer roles starting 2026. What I've been building: ▸ At Xelron — built and validated AI-agent benchmark scenarios across Terminal-Bench, Finch, and Terminus; designed 30+ deterministic, oracle-validated edge-case tests; analyzed agent execution traces to find failure patterns and fed the fixes back into benchmark design. ▸ CodeSense — a full-stack AI code review platform that diagnoses bugs, generates fixes, and renders a Monaco diff view, with a two-layer evaluation pipeline (LLM-as-judge + rule-based checks for accuracy, clarity, completeness, and hallucination risk). Next.js, TypeScript, PostgreSQL, Prisma. ▸ Sentinel — a privacy-preserving federated reinforcement learning system (PPO/A2C, PyTorch) that detects and mitigates network anomalies across distributed gateways without ever sharing raw data. Toolkit: Python, JavaScript/TypeScript, C++, SQL | LangChain, OpenAI Agents SDK, RAG, vector DBs, PyTorch | FastAPI, Node.js, Express, Socket.io | React, Next.js, Tailwind | AWS, MongoDB, PostgreSQL, Prisma, Docker Fundamentals matter to me: 500+ DSA problems on LeetCode, Codeforces Specialist (max 1401), and regular contests on CodeChef and GeeksforGeeks. Open to full-time SDE and AI Engineer opportunities in Bengaluru, Hyderabad, or remote. Fastest way to reach me is ayushyadav893972@gmail.com — I reply to every message.

Location

Greater Bengaluru Area

AI stack

antigravity

claude code

claude

codex

cursor

gemini

Work experience

Software Engineer Intern at Xelron, Feb 2026 - Jul 2026 · 6 mos, • Built and validated AI-agent benchmark scenarios (debugging, CLI execution, error handling) across Terminal-Bench, Finch, and Terminus, improving evaluation coverage and reliability for LLM-based coding agents. • Designed 30+ deterministic, oracle-validated edge-case tests for terminal-based agent tasks, tightening the signal-to-noise ratio of automated evaluation pipelines. • Analyzed AI agent execution traces to identify failure patterns and evaluation inconsistencies, feeding fixes back into benchmark design. • Collaborated with senior engineers to refine automated evaluation pipelines, task specs, and validation logic for scalable agent benchmarking. Stack: Python, Linux/CLI, Docker, Git, automated & oracle-based testing

Education

Bachelor of Technology, Computer Engineering, Motilal Nehru National Institute of Technology

Class 12th, Secondary Education and Teaching, Beena Public School, Varanasi

Public links

github: https://github.com/Ayushy21