AI & Machine Learning Engineer

Mohammad Ashad Khan

I build machine-learning systems you can actually trust.

MSc Artificial Intelligence at Milano-Bicocca. I pair strong computer-vision and deep-learning models with explainability, so every prediction comes with the reason behind it. Currently building Lagani, an explainable stock-risk platform for Nepal's NEPSE market.

View projects → Resume GitHub

Based in Milan, Italy Open to AI / ML & CV roles EN · NE · HI · UR · IT

Shipped projects

Certifications

Years building

Published paper

01 — About

The why behind the work

I'm a Master's student in AI for Science & Technology at the University of Milano-Bicocca, with a B.Tech in Computer Science from Lovely Professional University. My work sits at the intersection of computer vision, deep learning, and explainable AI.

The thread through everything I build is trust. A model that predicts "high risk" or "defective" is useless if no one understands why — so I pair strong models (CNNs, Vision Transformers, gradient boosting) with explainability methods (Grad-CAM, LIME, SHAP, saliency, knowledge graphs).

I'm also a full-stack developer (React, Next.js, Supabase). I don't stop at a notebook — I ship models as live, interactive products people can use.

02 — Toolkit

Skills & technologies

The stack I reach for, from research notebooks to production web apps.

AI / Machine Learning

PyTorchDeep LearningComputer Vision CNNsVision TransformersNLP Transfer LearningGradient BoostingScikit-learn SBERT1D CNN

Explainable AI

Grad-CAMLIMESHAP Saliency Mapst-SNEKnowledge Graphs TransE Embeddings

Vision & HCI

OpenCVMediaPipeGesture Control Depth EstimationPygameImage Classification

Web / Full-Stack

ReactNext.jsTypeScript Node.jsExpressTailwind CSS Streamlit

Data & Infra

PostgreSQLMongoDBSupabase Neo4jPandasNumPy

Languages & Tools

PythonJavaScriptSQL GitDockerHTML / CSS

03 — Work

Selected projects

Research-grade ML with real metrics — and full-stack products that ship.

Flagship · In progress

Lagani

Explainable AI platform for NEPSE stock-risk analysis

80.9%

Test acc

0.81

Macro-F1

120

NEPSE stocks

Leak-free, symbol-disjoint risk model (gradient boosting + MLP) on 120 Nepali stocks
CNN trend model, volatility-regime classifier & TransE knowledge-graph embeddings
LIME + saliency explanations and a Streamlit dashboard for transparent risk scoring

PyTorchGradient BoostingLIMENeo4jStreamlit

↗ Code Live demo soon

Lagani system pipeline diagram — **System pipeline**

Feature importance chart — **Global XAI** · feature importance

LIME local explanation for NABIL — **Local XAI · LIME** · NABIL

Vision Transformer · XAI

FreshGuard

AI fruit-quality inspection with explainable predictions

Vision Transformer transfer-learning on FruitNet for fresh-vs-defective classification
LIME super-pixel explanations show which part of the fruit drove each decision
t-SNE projection of learned embeddings to visualise class separation

ViTPyTorchTransfer LearningLIMEt-SNE

↗ Code

Vision Transformer · XAI

Bloom XAI

Explainable flower classification with ViT transfer learning

Fine-tuned a pretrained Vision Transformer on a multi-class flower dataset
LIME local explanations + t-SNE embedding visualisation for interpretability
Built as the AIML exam deliverable with a full report & reproducible notebook

ViTPyTorchLIMEt-SNE

↗ Code

From-scratch CNN · XAI

CIFAR-10 CNN + XAI

A custom CNN benchmarked against three explainability methods

Designed and trained a convolutional network from scratch on CIFAR-10
Compared Grad-CAM, saliency maps and LIME to understand model attention
Single reproducible Colab notebook — model, training & explanations

PyTorchGrad-CAMSaliencyLIME

↗ Code

Deep Learning · Healthcare

CardioNet

Deep-learning ECG arrhythmia detection

Custom 1D CNN on the MIT-BIH dataset for arrhythmia classification
Interactive Streamlit dashboard for ECG visualisation & real-time prediction
Evaluated with accuracy, precision, recall, F1 & confusion matrix

PyTorch1D CNNStreamlitScikit-learn

↗ GitHub

Computer Vision · HCI

GestureRun

Gesture-controlled endless runner

Real-time webcam hand-gesture recognition for lane switching, jumping & sliding
Built for the HCI & Intelligent Consumer Technologies course
Runs in real time on a standard webcam via MediaPipe Hands

OpenCVMediaPipePygamePython

↗ GitHub

NLP · Explainable Matching

Resume & Job Match Scorer

Semantic job matching with explainable AI

SBERT semantic similarity between resumes and job descriptions
LIME explanations surface which skills drove each match score
Neo4j skills knowledge graph for structured reasoning

SBERTLIMENeo4jPython

↗ Code

Full-Stack · AI

Dashvoard

AI productivity dashboard

AI auto-categorisation parses URL metadata into Study / Work / Social / Shopping
Granular per-link and per-folder privacy toggles
React + Supabase real-time backend

ReactTailwindSupabase

↗ Code

Full-Stack · FinTech

EchoBook

Multi-ledger finance manager

Partitioned ledger separating Personal, Business & Lending accounts
Real-time balance, credit & debit metrics
Typed React + Supabase architecture

ReactTypeScriptTailwindSupabase

↗ Code

04 — Momentum

What I'm working on & writing

Currently

Adding SHAP & counterfactual explanations to Lagani's risk model
Exploring faithfulness of XAI methods as a thesis direction
Shipping live demos of my Streamlit projects to the web
Learning Italian (A1 → A2)

Writing

ICCS-2023 "KILBY100" — Conference Publication

Peer-reviewed · ICCS 2023

Published

Why LIME explanations can be unstable — and what to do about it

Deep-dive on local-surrogate reliability

Draft

Explaining a Vision Transformer's decisions on fruit quality

From attention maps to LIME on FreshGuard

Draft

05 — Journey

Experience & education

Experience

May 2022 – Jun 2025

Frontend Web Developer

Securedsoft · Remote

Built responsive web interfaces with React, Tailwind & Supabase APIs; improved performance and cross-browser compatibility.

Apr 2024 – Feb 2025

Computer Science Instructor

Lumbini World School, Nepal

Taught programming fundamentals to K-12 students; delivered the Nepal STEM Alliance-certified Coding ToT curriculum.

Jan 2020 – May 2023

Management Team Member

Google Developer Student Club, LPU

Organised hackathons & workshops; mentored peers in software development.

Education

MSc Artificial Intelligence for Science & Technology

Università degli Studi di Milano-Bicocca, Italy

Computer Vision, Deep Learning, Explainable AI, Big Data & Signal processing.

B.Tech Computer Science Engineering

Lovely Professional University, India

Foundations in CS, software engineering & a published ICCS-2023 conference paper.

Certifications

ICCS-2023 "KILBY100"

Conference Publication · May 2023

Neo4j Certified Professional

Jun 2025

AI Engineer for Developers Associate

DataCamp · Mar 2026

AI Fundamentals

DataCamp · Mar 2026

NASA ARSET Remote Sensing

Fundamentals · Nov 2025

HP LIFE Design Thinking

Nov 2024

Let's build something explainable.

Open to AI / ML & Computer Vision roles, internships, and research collaborations. Drop a line and it lands with me straight away.

LinkedIn GitHub Resume