FNB

I'm Fardeen NB

AI Engineer who builds production LLM systems and trains models from scratch. I have shipped six AI features into a hospital's revenue cycle, and I pre-trained Neutrino-Instruct, an open 7B-parameter language model.

Biography

I started out as a builder. In 2019, as a teenager, I founded FX Pattern Pro, a desktop application that helps quantitative traders detect algorithmic chart patterns. Shipping it taught me how to turn noisy, high-volume market data into software people could trust.

That work pulled me into machine learning. At Accenture, I fine-tuned enterprise language models and built production retrieval and evaluation pipelines. I then went deeper into the fundamentals with a Master of Engineering in Artificial Intelligence at the University of Cincinnati, where my research on attention and long-range dependencies in transformers led to peer-reviewed publications.

Today I build production AI at Insight Hospital and Medical Center, where I have shipped six AI features across the hospital's revenue cycle, from medical coding to claims appeals and patient calls. I also lead NeuralCrew, my independent research effort, where I pre-trained and post-trained Neutrino-Instruct, a 7-billion-parameter language model, from scratch to test how far careful curation of open data can go without frontier-scale compute.

Work Experience

Insight Hospital and Medical Center

Jun 2025 - Present

Full Stack AI Engineer

Chicago, Illinois · Hybrid

Owned and shipped 6 production AI features across the hospital's Revenue Cycle Management (RCM) platform, automating end-to-end medical coding, claims appeals, prior authorization, and patient communications.
Fine-tuned specialized LLMs to extract and predict ICD-10 and CPT medical codes directly from clinical documentation, implementing a human-in-the-loop validation pipeline for certified medical coders prior to claim submission.
Built a RAG-powered claims appeal agent that synthesizes payer coverage policies with EDI 837 and 835 claim transaction files to draft denial appeal letters, automatically routing low-confidence drafts to human reviewers.
Developed HIPAA-compliant autonomous agents for prior authorization submissions and payer claim follow-ups, replacing two manual telephone workflows with strict confidence thresholds and validation guardrails.
Designed a real-time inbound call-routing pipeline (STT, LLM, and TTS over RingCentral) that classifies patient intent and routes inquiries to appropriate clinical departments without operator intervention.
Engineered conversational voice and SMS agents with ultra-low response latency to automate appointment reminders and patient communications, eliminating repetitive front-desk phone outreach.
Developed core frontend and backend services across the hospital's broader RCM infrastructure using modern full-stack web technologies.

Accenture

Jun 2022 - Aug 2024

Artificial Intelligence Engineer

India · Hybrid

Fine-tuned enterprise large language models (LLMs) for domain-specific knowledge extraction, document summarization, and semantic search across client data lakes.
Engineered production Retrieval-Augmented Generation (RAG) pipelines integrating scalable vector databases, hybrid sparse-dense retrieval, and semantic reranking.
Built comprehensive LLM evaluation and benchmarking frameworks measuring inference accuracy, factual consistency, hallucination rates, and system latency in production.

Education

University of Cincinnati

Aug 2024 - May 2026

Master of Engineering, Artificial Intelligence

Cincinnati, OH

Specialized in foundational deep learning, transformer mechanics, attention optimization, and distributed neural network pre-training.
Completed advanced graduate coursework in Deep Learning, Reinforcement Learning, Natural Language Processing, Computer Vision, and Convex Optimization.
Conducted graduate research on attention distance bottlenecks and long-range sequence dependencies in transformer architectures, leading to peer-reviewed publications.

Skills & Technologies

PyTorch

Hugging Face

OpenAI

JAX

Triton

CUDA

DeepSpeed

vLLM

Python

C++

Git

Docker

Kubernetes

Typescript

React

Next.js

Node.js

Postgres

Certifications

Certified Artificial Intelligence Scientist

Jun 2025

United States Artificial Intelligence Institute

Validated competencies across deep learning architectures, predictive modeling, generative systems, and enterprise AI lifecycle engineering.
Demonstrated mastery in architecting scalable machine learning pipelines using TensorFlow, Keras, and neural optimization methodologies.

NVIDIA-Certified Associate: Generative AI LLMs

Apr 2025

NVIDIA

Certified in large language model foundations, transformer mechanics, attention dynamics, and parameter-efficient fine-tuning (PEFT/LoRA).
Validated proficiency in designing high-throughput inference pipelines, prompt engineering strategies, and production RAG workflows using NVIDIA accelerated compute.

NVIDIA-Certified Associate: Generative AI Multimodal

May 2025

NVIDIA

Certified in multimodal architecture design, vision-language model (VLM) integration, cross-attention alignment, and diffusion generative paradigms.
Demonstrated capabilities in constructing multi-sensor data pipelines, optimizing cross-modal embeddings, and deploying accelerated inference on NVIDIA systems.

Pre-trained LLM

Neutrino-Instruct

7B instruction-tuned language model · Pre-trained from scratch

Hugging Face

Neutrino-Instruct is a 7-billion-parameter language model that I pre-trained and post-trained from scratch. The project tests how far careful curation of open data can go without frontier-scale compute. It uses a custom architecture tuned for multi-turn conversation and instruction following, and it runs locally through GGUF and Ollama.

Model Specifications

Parameters: 7 Billion
Context Length: 32,768 tokens
Formats: GGUF · Safetensors
License: Apache-2.0
Architecture: Custom (Neutrino)
Language: English

Quick Start (Local Inference)

Ollama CLI

ollama run fardeen0424/neutrino

Training Datasets (13 Curated Sources)

FineWeb-Edu

FineWiki

Wikipedia (En)

FinePDFs

GitHub Code

TinyStories

Awesome ChatGPT Prompts

SlimPajama

Cosmopedia

RedPajama-V2

Dolma

Stack-v2

UltraChat

Download Analytics

Totals from the Hugging Face API. The daily curve is an estimate.

Publications & Research

Google Scholar Citations

Citations per year

Total Citations302

h-index12

Get in Touch

Open to AI rolesOpen to full-time AI engineering roles

Let's build the next generation of AI systems.

I build production LLM systems, from retrieval pipelines and agents to models trained from scratch. I'm looking for AI engineering roles where deep technical work ships to real users.

Full-Stack AI Engineering

RAG & AI Agents

LLM Inference & Optimization

Foundation Model Pre-training

Distributed Training (DeepSpeed, CUDA)

Connect on LinkedIn

Cincinnati, OH · Open to relocation