01
End-to-end delivery
I own the full model lifecycle — preprocessing, training, evaluation, Git-based delivery. Clean, documented, reproducible pipelines you can actually trust in production.
hello, world — ML engineer, lahore
Building AI that ships from data to deployed models.
ABDULLAH JAN — Electronics & Computing @ COMSATS Lahore. Five end-to-end ML projects built in real internships, from BERT fine-tuning to multimodal regression. Python · C++ · ML.
$ cat about_me.txt
Electronics & Computing undergraduate at COMSATS Lahore.
Focused on AI & machine learning — built and deployed real
ML projects during my internship at DevelopersHub, and now
working as an ML Engineer Intern at FlyRank AI.
stack: [python, tensorflow, scikit-learn, huggingface, opencv]
also: electronics — from a line-following robot to a portable
function generator, software meets hardware.
$ ▌
scroll
about
01
From coursework to deployed results — one engineering loop.
I'm an Electronics & Computing undergraduate specializing in AI and machine learning. I've earned hands-on experience across NLP, computer vision, and tabular ML through a real internship where I built five end-to-end ML projects — all public on GitHub. I enjoy working where data meets real-world problems — predictive models, NLP, or full ML pipelines. Ask me anything →
02
Deep learning & NLP
BERT fine-tuning, transformers, LSTM forecasting, U-Net segmentation, YOLO detection, and multimodal networks — trained and evaluated with a focus on real metrics, not just notebooks.
03
Electronics & hardware
From a line-following robot to a portable function generator — building electronics keeps me grounded in how software meets hardware, end to end.
python : lang
tensorflow : deep-learning
scikit-learn : ml
huggingface : nlp
pytorch : dl
pandas / numpy : data
opencv : vision
u-net / yolo : segmentation
c++ : lang
git / github : tools
linux : shell
by the numbers
02
Results, proven on real projects.
process
03
How I build — data to delivery.
I follow four key steps to be confident the model ships and holds up in the real world.
01
Understand the problem
Define the target metric, explore the data, and map business constraints to a modeling approach.
02
Prepare the data
Feature engineering, cleaning, oversampling (SMOTE), and train/val splits that reflect reality.
03
Train & evaluate
Pick the right architecture, train with proper metrics, and compare baselines honestly.
04
Ship & document
Package results, write up findings, and deliver through Git so anyone can reproduce the run.
projects
04
Selected work — each one end-to-end.
// NLP · BERT
BERT News Classification
Fine-tuned bert-base-uncased → 4 news classes
- 01Hugging Face datasets + Trainer API
- 02Class-imbalance handled explicitly
- 03Eval vs. baselines on held-out split
// ML PIPELINE
Telco Customer Churn Pipeline
SMOTE + XGBoost → predict who will cancel
- 01Feature engineering pipeline
- 02SMOTE oversampling for imbalance
- 03Cleaning → training → eval automated
// MULTIMODAL
Multimodal House Price Prediction
CNN visuals + MLP tabular → late-fused regression
- 01Dual-branch network design
- 02Structured + image feature fusion
- 03Regression head on fused latents
// LLM · NLP
AI Health Query Chatbot
Intent classification + guardrails → safe answers
- 01Llama-3-8B via HF Inference API
- 02Rule-augmented response generation
- 03Emergency-detection guardrails + UI
// TIME SERIES
Stock Price Forecasting
LSTM sliding window → next-day close
- 01Sliding-window preprocessing
- 02Historical market data via yfinance
- 03Random Forest baseline compared
// CV · SEGMENTATION
Image Segmentation · U-Net
Encoder–decoder with skip connections
- 01U-Net from scratch (Dice loss, IoU)
- 02Semantic segmentation evaluation
- 03Metrics-driven prediction review
also on GitHub: strange-portals AR (OpenCV + MediaPipe) · heart-disease prediction · iris EDA — github.com/JAN-tech404 ↗
case studies
05
The numbers behind the work.
// BERT · NLP
2026 · DevelopersHub
News classification — accuracy
Fine-tuned bert-base-uncased on 4 classes. Class-imbalance handled with weighted loss. Final accuracy 89.60% on the held-out split.
Open case →
// LSTM · TIME SERIES
2026 · DevelopersHub
Stock forecast — next-day close
LSTM with sliding-window preprocessing forecast next-day closes. Random Forest baseline compared across the same window. Predictions tracked against actuals.
Open case →
// U-NET · CV
2026 · ITU Program
Semantic segmentation — training loss
U-Net trained with Dice loss + IoU. Loss converges steadily across epochs; segmentation maps evaluated per-class with standard metrics.
Open case →experience
06
Where I've shipped — internships & programs.
-
June 2026 — Present
Machine Learning Engineer Intern — FlyRank AI
Working through practical ML engineering tasks spanning Python, deep learning, computer vision, segmentation, object detection, and transformers. Building and refining data-prep → training → evaluation workflows with notebook-based, Git-documented delivery.
-
Feb 6 — Mar 27, 2026 · Best Award · DHC-1821
AI/Machine Learning Intern — DevelopersHub Corporation
Completed a 6-week internship with exceptional performance, building five end-to-end ML projects independently:
· Fine-tuned BERT for multi-class news classification (Hugging Face Transformers) · built an end-to-end telco churn pipeline with feature engineering, SMOTE oversampling & XGBoost · developed a multimodal house-price predictor (tabular + property images, dual-branch NN) · deployed an AI health-query chatbot with intent classification · forecasted stock prices with LSTM time-series models.
-
Mid 2026
AI/ML Intern — ITU Internship
Six-week hands-on program: Python fundamentals → NumPy/Pandas/OpenCV → Machine & Deep Learning → Semantic Segmentation & U-Net → Object Detection & YOLO → Transformers & Hugging Face.
education & achievements
07
// BS Electronics & Computing
BSEC — COMSATS University Islamabad
Lahore Campus · Sep 2024 — Sep 2028Specializing in AI/ML — programming, electronics & computing, mathematics, and AI coursework. Available for entry-level ML/AI roles.
// Intermediate · Pre-Engineering
Punjab Group of Colleges
Oct 2022 — Aug 2024Pre-engineering foundation — mathematics, physics, and chemistry groundwork for engineering studies.
// Best Award
6-Week AI/ML Internship · DHC-1821
DevelopersHub Corporation · Feb–Mar 2026Recognized with exceptional performance for building five end-to-end ML projects during the internship program.
certifications
08
Proof of work — certificates.
DevelopersHub — 6-Week AI/ML Internship · Best Award
Certificates collage — all certifications
Claude platform certificates
IBM SkillsBuild — Foundations of AI
IBM SkillsBuild — RAG with LlamaIndex
Anthropic — Claude 101
// more certifications
IBM SkillsBuild (Jul 2026) and a full Anthropic Claude fluency series:
Foundations of AI
RAG with LlamaIndex
AI Fluency for Builders
AI Fluency for Educators
AI Fluency for Students
AI Fluency for Non-profits
AI Fluency for Small Business
AI Fluency Cap & Limits
AI Fluency Framework
Teaching AI Fluency
Claude 101
Claude Code 101
Claude Code in Action
Claude Platform 101
Claude with Amazon Bedrock
Claude with Anthropic API
Claude with Google Vertex AI
Intro to MCP
MCP Advanced Topics
Intro to Agent Skills
Intro to Subagents
Intro to Claude Cowork
partners
09
Trusted by — the people I work with.
FRFlyRank AI
DHDevelopersHub
ITUITU Internship
CUCOMSATS Lahore
PBCPunjab Group
get in touch
10
Let's run the numbers on your problem.
Send me your data and constraints — I'll come back with an approach and the expected impact. I reply within a day.