Headline

Final Year AI Undergrad | Building Production LLM Systems | GenAI · RAG · FastAPI · PyTorch

About

I am an AI & Data Science student focused on LLM safety, evaluation, and reliability the question of how we actually know a model is behaving as intended. Most of what I know I learned by building and stress-testing systems, not just reading about them. What I have built and shipped over the past year: 🔹 FIE — an open-source LLM guardrail (PyPI SDK, 7,500+ downloads, deployed on GCP) that flags prompt injection and jailbreaks on CPU in ~48 ms. While benchmarking it, I found that ~52% of JailbreakBench had leaked into common training data, built a decontamination pipeline across 8 benchmarks, and re-reported an honest held-out recall of 86%. Documented in two technical papers (with DOIs). 🔹 ARIA — an agent-reliability framework that diagnoses why an agent fails, not just whether it did (91.2% failure-detection precision on GAIA Level 1). 🔹 A dyslexia early-screening system (ResNet-50 + OCR + Grad-CAM, 84.9% ROC-AUC) and a multilingual mental-health assistant supporting 14 Indian languages with RAG and voice. These aren't separate side projects — they are one direction I care about: making LLM systems reliable end to end. Detect when a model fails (FIE), diagnose why (ARIA), and prevent bad outputs before they reach a user. I enjoy working across the full AI stack model development and evaluation, APIs, and deployment. Tech: Python, PyTorch, FastAPI, LangChain/LangGraph, RAG, FAISS, Docker, GCP, React, MongoDB Currently looking for GenAI / AI Engineer roles where I can build impactful products, learn from strong teams, and contribute from day one.

Location

Pune District

AI stack

antigravity

claude code

hugging face

langchain

langgraph

Guardrails AI

Work experience

Trainee at Samsung Innovation Campus, Dec 2025 - Feb 2026 · 3 mos, Pune City, Maharashtra, India, Completed advanced Python training with practical projects and framework-based development. Built a Smart Mental Wellbeing Assistant during the program's hackathon, applying AI concepts to solve real-world challenges.

Summer Internship at CollaMeta, May 2025 - Jul 2025 · 3 mos, Remote, As an Android Developer Intern at Collameta, I worked with a team to build a mobile app focused on enhancing the cognitive wellness of senior citizens through interactive assessments and brain-training games.

Education

Bachelor of Technology - BTech, Computer Science (AI&DS), Pimpri Chinchwad University, Pune

Public links

github: https://github.com/AyushSingh110

medium: https://medium.com/@ayushsingh355vns