Looking for a job

Ayush Parwal

@ayushparwal1

Message

ML engineering intern at Grafyn AI who built pipelines to detect adversarial attacks and harden production agents.

India

Message

At ISF, I constructed a backend pipeline to turn unstructured LLM outputs into schema-validated data using Pydantic. I also deployed FastAPI endpoints with fallback and error-recovery logic for reliable model inference.

At Grafyn AI, I designed ML security pipelines on AWS, Databricks, and Snowflake to simulate and detect model-stealing, perturbation, and data-poisoning attacks. I also collaborated with Red Teaming to harden production agents against jailbreaking, prompt injection, obfuscation, and LLM hijacking.

I built Cavora, a conversational AI platform using React.js and TypeScript, and shipped it to production on cavora.tech. It supports real-time interactions for 45+ users and includes a Meta LLaMA chatbot with Google Search API integration.

My projects also include a CrewAI-based deep research assistant and a Secure RAG system with safeguards against prompt injection, retrieval poisoning, and data leakage. I pioneered tantheta, an open-source Python library for complex calculations in math, chemistry, and physics, and published it on PyPI.

Experience

Work history, roles, and key accomplishments

Full Stack AI/ML Intern

ISF

Jan 2026 - Jun 2026 (5 months)

Constructed a backend pipeline to parse unstructured LLM outputs into structured, schema-validated data using Pydantic, and deployed FastAPI endpoints with fallback/error-recovery logic to integrate model inference reliably across front-end and backend systems.

Pydantic fastAPI LLM Backend Development API Integration

Founding ML Engineering Intern

Grafyn AI

Jul 2025 - Dec 2025 (5 months)

Designed and scaled ML security pipelines on AWS, Databricks, and Snowflake to simulate and detect model stealing, perturbation, and data poisoning attacks, improving detection coverage of adversarial ML threats. Collaborated with Red Teaming to harden production agents against jailbreaking, prompt injection, obfuscation, and LLM hijacking using automated testing workflows.