amit = {
    "name"      : "Amit Dubey",
    "location"  : "India 🇮🇳",
    "focus"     : "AI / ML Engineering",
    "philosophy": "Build real systems, not just train models.",
    "currently" : ["DSA", "MLOps & Deployment", "Transformer internals"],
    "goal"      : "Production-grade AI that works outside notebooks."
}

• 🔭 I focus on end-to-end AI pipelines — from data to deployment
• 🧠 I care about reliability more than raw model performance
• 🛠️ I think in terms of production systems, not just notebooks
• 🔁 I'm comfortable failing and iterating until something actually works

Languages

ML / Data

AI / LLM

Tools

Built an LLM-based agent for multi-step reasoning tasks inspired by GAIA-style problems. Combines structured execution with reasoning to handle complex queries reliably.

Key learnings: LLMs need control · System design > model choice · Reliability > raw capability

Retrieval-augmented chatbot using embeddings and vector search to answer medical queries from a curated knowledge base — minimizing hallucination over pure generation.

Key learnings: Retrieval quality defines everything · Chunking strategy matters · LLM eval is hard

Structured dataset for nutrition-based AI queries, usable for training or evaluation in food/health AI systems.

Key learnings: Data quality > model quality · Building datasets is harder than using them

Deep learning pipeline for aneurysm detection on real-world noisy medical imaging data from the RSNA Kaggle competition.

Key learnings: Real data is messy · Debugging is a core skill · Pipelines break more than models

"I prefer understanding systems deeply rather than just using tools."

"Reliability matters more than raw capability."

"Production-grade AI works outside notebooks and scales reliably."

Open to collaborations, internships, and interesting AI/ML problems.