Initializing...
faizy.dev
Home About Me Projects GitHub Certifications Blog Contact
Mohd Faizy - AI/ML Engineer
ai-assistant.sh
$
$
$

AI/ML Engineer

class AIArchitect :
def __init__ (self):
self.stack = [ 'Transformers' , 'LLMs' , 'AGI' ]
def innovate (self): return f"🚀 {self.stack} → Future"

Electronics Engineer specializing in Neural Networks, Machine Learning, and Deep Learning. From VLSI and microelectronics to computer vision and NLP, I design intelligent systems that drive the next wave of technology. #LLMs #GenAI #AgenticAI

Explore My AI Projects Download Neural Resume
0
ML Projects
0
AI Certifications
0
Years Experience
0
GitHub Commits

Discovery Stream

Curating the latest breakthroughs in AI, technology, and science that shape our future.

Live Updates
AI-Powered Curation
FEATURED
AI Breakthrough Latest Snapshot

GPT-5.4 & Gemini 3.1: The Agentic Frontier

The 2026 frontier models represent a leap far beyond basic reasoning. With infinite context capabilities and native vision-language-action (VLA) execution, these systems autonomously govern complex software ecosystems and pioneer independent scientific research.

Explore Research
Trending Share
Architecture Context Memory

Infini-attention (Google)

Leave the million-token limit behind. This technique introduces a compressive memory algorithm into the standard attention mechanism, allowing Transformers to process infinitely long context sequences with bounded memory and compute hardware footprints.

Read Deep Dive
Quantization 1-bit Era

BitNet b1.58 (Microsoft)

The end of Matrix Multiplication in LLMs. A ternary weight parameterization (-1, 0, 1) mapping that achieves identical performance to extreme 16-bit precision models, while slashing memory needs and latency exponentially for purely arithmetic-free inference.

Read Deep Dive
Dynamic Router DeepMind

Mixture-of-Depths (MoD)

Why calculate everything? MoD allows large language models to dynamically allocate compute based on individual token complexity. Easy tokens skip deeper transformer layers entirely, slashing total FLOPs without hurting downstream benchmark scores.

Read Deep Dive
Reasoning Silent Logic

Quiet-STaR Learning

Teaching language models to 'think' before speaking. Quiet-STaR augments LLMs with the ability to pause and generate implicit rationale sequences behind the scenes before deciding on the final output tokens—vastly improving logical coherence.

Read Deep Dive
Alignment Tuning

Direct Preference Optimization

Bypassing RLHF. DPO represents the shift away from brittle reinforced reward networks, substituting it with a closed-form algorithm that enforces human preference natively into the language model's policy using standard binary cross-entropy.

Read Deep Dive
Compute Opt Context

Ring Attention with Blockwise Compute

Solving the quadratic attention explosion by distributing long sequences across a ring of GPU hosts. This paper proves we can scale context windows over 10 million tokens by calculating self-attention block-by-block asynchronously.

Read Deep Dive
Hybrid Arch SSMs

Jamba: Joint SSM + Transformer

A novel architecture mixing Mamba (State Space Model) layers with traditional Transformer blocks. This breakthrough allows production-ready enormous contexts with near-instant caching and linear scaling, merging the best of two worlds.

Read Deep Dive
Self-Alignment RLHF-Alternative

Self-Rewarding Language Models

Shattering the human limitation barrier. This Meta research shows how advanced LLMs act as their own reward judges, perpetually evaluating and refining their outputs, outstripping the quality of datasets entirely produced by humans.

Read Deep Dive
Inference Acceleration

Speculative Decoding

A paradigm shift for massive scale inference latency. By linking a fast 'draft' model to speculatively compose sequences for a massive 'oracle' model to verify in parallel, LLMs can output text 2-3x faster with mathematically strictly zero accuracy loss.

Read Deep Dive
RNN Revival Linear Scaling

RWKV (Linear RNNs)

Receptance Weighted Key Value network scaling. RWKV combines the highly parallelizable training mechanics of a Transformer with the highly efficient O(1) inference structure of RNNs, representing one of the best open-weight linear attempts online.

Read Deep Dive
Swipe to explore breakthroughs
View All Discoveries

About Me

AI/ML Engineer · Self-taught · Research-driven · Based in New Delhi

Personal Journey Available

Who am I?

I'm a Data Science enthusiast holding an M.Tech in Electronics & Communication Engineering from Jamia Millia Islamia University, New Delhi.

As a self-taught programmer, I specialise in NLP, Large Language Models, Agentic AI, and Retrieval-Augmented Generation (RAG).

Technologies Studied & Applied
LLM Research
Transformers Infini-Attention Quiet-STaR RWKV Jamba SSM Spec. Decoding
NLP & Generative AI
BERT GPT-2 (scratch) DPO / RLHF LangChain LangGraph RAG
Computer Vision
YOLOv3 Siamese Nets Medical Imaging Emotion AI
Neural Networks
AI/ML Engineer
Self-taught
50+
Projects
5+
Years
15+
Certs
1.2K+
Commits/Yr
M.Tech

Electronics & Communication Engineering

Jamia Millia Islamia, New Delhi

Master of Technology · 2017 – 2019

Specialisation in VLSI Design 8.6 CGPA
B.E

Electronics & Communication Engineering

Jamia Millia Islamia, New Delhi

Bachelor of Engineering · 2012 – 2016

First Division Distinction 72%
Diploma

Electronics & Communication Engineering

Jamia Millia Islamia, New Delhi

Diploma · 2009 – 2011

The origin of my "Silicon to Software" journey. This rigorous hands-on hardware training cultivated my systems-level thinking, forming the bedrock for my current work in deploying highly-optimized neural networks on edge devices.

Hardware Foundations
C
First Division with Honors 84%
Academic Projects

Hardware & VLSI Background

  • IC Routing Optimization using Fully Connected Neural Networks (M.Tech) Developed an innovative approach to integrated circuit routing using deep learning techniques. The system achieved state-of-the-art performance in routing optimization.
  • Floating Point Multiplier & Simulation - VHDL ISE™ (B.E.) Designed and implemented a high-performance floating-point multiplier for digital signal processing applications. The project focused on FPGA implementation and optimization.
  • PIC Controller Microcontroller Based Electronic Lock System (Diploma) Developed a password-protected electronic lock system using PIC16F877A microcontroller. Features include keypad interface, LCD display, and secure access control.

Core Tech Stack

Actively Used
Python
Python
PyTorch
PyTorch
TensorFlow
TF
🤗
HF
LangChain
LangGraph
Docker
MLflow
Currently exploring
Diffusion Models Responsible AI Multimodal LLMs

Projects

Showcasing 50+ AI/ML projects across various domains - from production AI systems to research implementations. Featured below are my latest production-ready projects.

Deep Learning
AI Systems
Production Ready

More Projects

YOLOv3 Detection

Real-time Object Detection

COVID-19 Detection

Medical Imaging AI

Emotion AI

Facial Analysis

GPT-2 Training

From Scratch

EN-FR Translator

Neural Translation

BERT Sentiment

Advanced NLP

Linear Regression

NumPy Implementation

Siamese Networks

Similarity Learning

Anomaly Detection

Time Series

Plotly Dashboards

Interactive Viz

Geo Visualization

World Map Data

Pandemic Simulation

Epidemiology Model

8+
Production AI Systems
15+
Computer Vision
12+
NLP & GenAI
20+
ML & Data Science
Explore All 50+ Projects

GitHub

Open-source contributions, AI experiments, and collaborative projects. Building the future of AI, one commit at a time.

30+ Repositories
500+ Stars
Open Source

@mohd-faizy

AI/ML Engineer & Open Source Contributor

Repositories 76+
Total Stars 218+
Followers 60+
Contributions 225+
Visit GitHub

Commit Activity

1,247

Commits this year

Languages

Python 65%
Jupyter 20%
Pytorch 15%

Contribution Activity

GitHub Contribution Graph
Less
More

Featured Repositories

Probabilistic-Deep-Learning-with-TensorFlow

Public

Probabilistic Deep Learning finds its application in autonomous vehicles and medical diagnoses. This is an increasingly important area of deep learning that aims to quantify the noise and uncertain...

Jupyter Notebook TensorFlow
69 34
View Repository →

Developing-Large-Language-Models

Public

Dive into LLM development! Learn cutting-edge techniques behind models like OpenAI's GPT-4, Meta's LLaMA 2, Mistral-7B, and Anthropic's Claude. Master PyTorch, build transformers, & fine-tune pre-t...

Jupyter Notebook LLM
1 0
View Repository →

GPT1-From-Scratch

Public

This project implements GPT-1 using PyTorch, focusing on foundational transformer architectures for natural language processing tasks.

Python GPT
0 0
View Repository →

Natural_Language_Processing_in_Python

Public

This repository contains the code and resources for the "Natural Language Processing in Python". This repository contains the core skills you need to convert unstructured data into valuable insight...

Jupyter Notebook NLP
1 0
View Repository →

Agentic_AI_using_LangGraph-_MCP

Public

Agentic AI framework built using LangGraph and Multi-Agent Control Plane (MCP) for building structured, goal-driven multi-agent systems.

Jupyter Notebook Agentic AI
0 0
View Repository →

GenAI-with-Langchain-and-Huggingface

Public

This repo is the comprehensive guide, covering Langchain integration with Huggingface models. Learn to build, deploy, and optimize cutting-edge AI applications through hands-on projects and real-wo...

Jupyter Notebook GenAI
0 0
View Repository →

Reinforcement_learning

Public

Exploring the fundamentals of reinforcement learning (RL) to build agents capable of navigating complex real-world environments and enhancing the training of large language models (LLMs).

Jupyter Notebook RL
0 0
View Repository →

PyTorch-Essentials

Public

Welcome to the PyTorch Essentials repository! This Repo aims to cover fundamental to advanced topics related to PyTorch, providing comprehensive resources for learning and mastering this powerful deep learning framework.

PyTorch Tutorial
3 1
View Repository →

LLMs_Applications_with_LangChain

Public

Build impactful LLM applications, including RAG workflows and agentic systems, using the LangChain framework!

Jupyter Notebook LangChain
0 0
View Repository →

Let's Collaborate!

Interested in contributing to AI/ML projects? Check out my repositories and let's build something amazing together.

Follow on GitHub Get in Touch

Certifications

Continuous learning through 15+ certifications from industry leaders.

15+
Certifications
5
Providers
2024
Latest

Deep Learning Specialization

Coursera • 2023

TensorFlow Developer

Google • 2023

Machine Learning

Stanford • 2022

AI for Everyone

Coursera • 2022

Python for Data Science

IBM • 2021

Data Analysis

Google • 2021

Computer Vision

DeepLearning.AI • 2021

Natural Language Processing

Coursera • 2020

SQL for Data Science

Coursera • 2020

AWS Machine Learning

Amazon • 2020

Algorithms Specialization

Stanford • 2019

MLOps Fundamentals

Google Cloud • 2024

Neural Networks & Deep Learning

DeepLearning.AI • 2019

Statistics for Data Science

Coursera • 2019

Docker & Kubernetes

Udemy • 2024

Coursera (8) Google (3) Stanford (2) DeepLearning.AI (2)
Insights & Intelligence

Read the Future.

Deep dives into AI architectures, model scaling, and production-grade MLOps deployment.

View Full Archive
Featured 12 min read

Understanding Transformers: The Architecture That Changed AI

A comprehensive deep dive into transformer architectures, attention mechanisms, and their revolutionary impact on natural language processing and multimodal models.

Read Deep Dive

Contact

Let's build something amazing together! Drop me a message and I'll get back to you soon.

Send Message

Message Sent Successfully!

Thank you for reaching out. I truly appreciate your interest and will get back to you shortly.

Connect With Me

GitHub LinkedIn Twitter/X Email

Location

New Delhi, India