Adeel Asghar
// Building AI Solutions with Code, Curiosity, and Data
// 01. about
About Me
AI/ML undergraduate (CGPA 3.85/4.00) specializing in applied machine learning, deep learning, and AI engineering. Proven record validated by multiple hackathon wins (1st place at HACKDATA V1, HITEC Mega Code War 2026), AI Engineering internships at CalderR and POF, and University of Macau research collaboration. GDG Tech Lead and founding member who has mentored 100+ students, with hands-on experience spanning computer vision, NLP/LLM systems, RAG pipelines, and full-stack AI deployment.
$ research --interests
// 02. education
Academic Journey
BS Artificial Intelligence
COMSATS University Islamabad, Wah Campus
🏆 Academic Excellence Scholar
// 03. achievements
Honors & Achievements
1st Place — HACKDATA V1 Hackathon
Team MadGroot · Sortd Project • 2026
Led core development for the Sortd Project, engineering high-performance automated solutions to claim the top place in a competitive multi-team hackathon.
1st Place — Pitch Perfect & AI Model Training Competitions
HITEC Mega Code War 2026
Won 1st Place in both Pitch Perfect and AI Model Training competitions, engineering predictive ML models under speed constraints and delivering a winning architecture pitch.
Academic Excellence Award
COMSATS University Wah Campus
Recognized for outstanding academic achievement, maintaining a top-tier CGPA of 3.85/4.00 within the BS Artificial Intelligence program.
1st Place — CS Quiz Competition
UET Taxila
Ranked first place in a high-intensity, multi-round technical tournament testing deep knowledge in computer science theory and algorithms.
1st Place — Speed Programming Competition
HITEC University
Engineered highly optimized and accurate algorithms under tight time controls, beating out regional university chapters.
Ranked 4th of 80+ Participants — Tech Quiz Competition
Tech Fest COMSATS Islamabad
Secured a top 4 position among 80+ competing teams, demonstrating broad and deep technical competencies in computer science topics.
// 04. skills
Tech Stack
Languages
LLM & RAG Systems
AI / Machine Learning
Deep Learning & Computer Vision
Data & Analytics
MLOps & Deployment
// 05. projects
Featured Projects
Sortd
🏆 Hackathon WinnerAI-powered content capture platform that bridges the 'Capture Gap' — automatically extracts, summarizes, and categorizes content from Instagram Reels, YouTube videos, and screenshots into organized, searchable knowledge. Built for HACKDATA V1 Hackathon.
💡 Key Insight
Multimodal AI pipeline combining Gemini Vision for screenshot OCR and Groq Whisper for audio transcription — zero manual effort from capture to categorized knowledge.
Facial Expression Recognition Research Collaboration
Deep learning research project in collaboration with the University of Macau. Validated the I-Center method across FER+ and RAF-DB benchmark datasets with exploratory testing on ExpW for facial emotion recognition.
💡 Key Insight
Achieved +3.89pp macro-F1 improvement over state-of-the-art baselines across a 10-seed test with 95% CI positive validation.
Multi-Agent Business Research Assistant
Production-grade multi-agent business intelligence platform featuring four specialized agents (Clarity, Research, Validator, Synthesis) collaborating in a LangGraph workflow. Features human-in-the-loop (HITL) interrupt states, Tavily search tool execution, and multi-turn session persistence via MemorySaver.
💡 Key Insight
Leveraging LangGraph checkpointers and conditional routing engines made pause-and-resume workflows during multi-turn sessions seamless to manage dynamically.
OptiMill
High-fidelity Manufacturing-as-a-Service (MaaS) Progressive Web App designed to bridge the gap between engineering designers and fabrication shops. Transforms complex 3D CAD assets (.stl, .step, .obj) into actionable manufacturing intelligence through geometric analysis and deterministic cost modeling.
💡 Key Insight
By offloading geometric calculations and Haversine geospatial calculations to FastAPI and caching results in Redis, discovery latency was minimized to under 45ms.
ThreatLens
Malware binary visualization and classification platform. Converts any binary file to a grayscale image using the Nataraj byte-to-image technique, then classifies it against 25 malware families using a fine-tuned ResNet50 model trained on the Malimg dataset.
💡 Key Insight
Malware classification is fundamentally a computer vision problem — different malware families produce visually distinct byte patterns. ResNet50 achieves 76.4% val accuracy across 25 families on the Malimg dataset.
DermVision
Full-stack AI medical application for skin lesion classification. Classifies dermoscopy images into 7 lesion categories using a fine-tuned MobileNetV2 model trained on HAM10000. Features JWT auth, drag-and-drop upload, risk assessment, and prediction history.
💡 Key Insight
Three-service architecture: React SPA → Node.js API Gateway → Django ML Engine running ONNX inference. MobileNetV2 fine-tuned in two phases achieves 78.4% val accuracy across 7 lesion classes.
BSL Hand Gesture Recognition
Deep learning system classifying 34 British Sign Language gestures with 95.74% accuracy and 95.78% F1-score. Uses MobileNetV2 transfer learning with CLAHE normalisation and MediaPipe hand landmark preprocessing. Trained on 34,000 balanced images and deployed as a live Streamlit web app.
💡 Key Insight
CLAHE normalisation on the LAB L-channel alone outperformed standard RGB preprocessing — preprocessing quality mattered more than model complexity at this scale.
Smart Recycle System
Industrial waste classifier using MobileNetV2 transfer learning achieving 86%+ accuracy across 12 material categories — outperforming a custom CNN baseline by 25%. Features real-time hazardous waste flagging with an efficient inference pipeline for industrial constraints.
💡 Key Insight
MobileNetV2 with fine-tuning outperformed a custom CNN baseline by 25% on the same dataset — transfer learning wins when data is limited and classes are visually similar.
Retail Sales Performance Analytics Dashboard
End-to-end ETL pipeline processing 500,000+ transaction records for a UK-based online retailer. Identified seasonal revenue spikes, top-performing product categories, and high-value customer segments through in-depth EDA. Visualised with interactive Tableau dashboards for business intelligence decisions.
💡 Key Insight
500K+ transaction records revealed that 80% of revenue came from just 3 product categories — classic Pareto, but only visible after cleaning 40% missing/duplicate records in the ETL stage.
// other projects
AI/ML Model Advisor
Rule-based expert system with custom forward and backward chaining inference engines that recommends optimal ML algorithms based on dataset characteristics. Includes explainability support and a lightweight Flask web interface for interactive user queries.
Indian Air Pollution Analysis
Analyzed historical air quality data (2015–2020) across 26 major Indian cities. Built an automated data pipeline with robust preprocessing, compared 4 regression models with hyperparameter tuning to forecast PM2.5 levels, and deployed an interactive Streamlit prediction dashboard.
Encrypted Monitoring App
Python application that periodically captures desktop screenshots, encrypts them using AES or DES, generates SHA-256 integrity hashes, and securely logs all data. Supports full decryption and hash verification.
Smart Farm Security System
Custom CNN built from scratch to classify 10 farm animal species and flag potential intruder threats from live image analysis. Trained on 28,000 images from the Animals-10 dataset and deployed via Streamlit for real-time inference.
CIFAR-10 Classical ML Classifier
CPU-efficient classical ML approach to CIFAR-10 using Histogram of Oriented Gradients (HOG) for feature extraction and SVMs for classification. Lightweight, interpretable, and runs without GPU.
Bring It Buddy
🧪 PrototypePrototype — a Laravel-based peer-to-peer delivery web app connecting senders with travellers going the same route for cost-effective package delivery.
// 06. experience
Work Experience
My professional journey building AI systems and leading technical communities.
100+
Students Mentored
2+
Years Building
AI Engineering Intern
CalderR
Jun 2026 – Present · Internship
- ▸Resolved a BM25 keyword-bias failure in a personal-document RAG assistant by implementing hybrid retrieval with cross-encoder reranking, reaching 14/15 correctly grounded Q&A responses across 444 chunks from 31 source documents
- ▸Built a real-time, multi-source research engine using a RAG pipeline with RAGAS-based relevance and faithfulness evaluation, generating automated reports beyond delivery requirements while hardening output formatting for reliability
- ▸Built four additional agentic AI tools — a CLI assistant, multi-agent research assistant, automated code-review agent, and data-analysis agent — covering LLM tool-calling and multi-step orchestration
AI Intern
Pakistan Ordnance Factories (POF)
Jul 2026 – Present · Internship
- ▸Joined the AI team to support applied machine learning workflows, including data preprocessing and model evaluation tasks
- ▸Ramping up on the team's production ML tooling and internal AI systems as part of an ongoing AI engineering rotation
Founding Member & Tech Lead
GDG On Campus — COMSATS Wah
Oct 2025 – Present · Community Leadership
- ▸Grew practical AI/ML skills for 100+ students by co-founding GDG On Campus at COMSATS Wah and organizing 5+ workshops and seminars on AI/ML, data science, and software development
- ▸Strengthened project teams' skills in Python, model training, and evaluation metrics (precision, recall, F1-score) by mentoring across supervised learning and computer vision tracks
- ▸Strengthened practical AI/ML skills for junior cohorts by delivering sessions on data preprocessing, feature engineering, and CNNs
DevOps & AI Intern
SPS — Software Productivity Strategists
Jul 2025 – Sep 2025 · Internship
- ▸Streamlined build-to-deploy workflows across development and staging environments by building CI/CD pipelines in Jenkins with automated triggers, test execution, and Slack/email notifications
- ▸Improved service scalability and uptime by containerizing services with Docker and Docker Swarm, maintaining health via container monitoring and restart policies
// 07. certifications
Certifications
Verified credentials from world-class institutions.
$ issuers --summary
Ligency, Ed Donner
AI Engineer Track — Core (LLM Engineering, RAG, QLoRA, Agents), Production & Agentic (MCP)
In Progress
DeepLearning.AI
Machine Learning Specialization (Supervised ML & Advanced Learning Algorithms)
Sep 2025
IBM
Data Analysis with Python
Sep 2025
Coursera
Python for Data Analysis: Pandas & NumPy
Jul 2025
LearnKartS
Git with GitLab & Bitbucket
Jul 2025
Harvard University
Introduction to Programming with Python (CS50P)
Aug 2024
// 08. contact
Let's Work
Together.
Open to freelance AI/ML projects, remote roles, and research collaborations. If you're working on a hard problem, I'd like to hear about it.