Loading Portfolio...
Hi there, I'm
I build search and retrieval systems, train the models behind them, and try to make it clear why a result showed up. Then I put that work into apps.
Years Experience
Projects Delivered
Publications
Models
Scroll to explore
Research papers on one side. Apps and models on the other.
I live in Seoul. I finished my MS in AI Convergence at Soongsil University (GPA 4.28/4.5) and now lead the software team at Ecoda. Before that I was a research assistant at Soongsil and built product work at Cublick. I worked in Pakistan for years, then moved here for the degree.
Most of my research is retrieval: finding the right document when the fields and layout keep changing. I train language models, build RAG pipelines, and care about explainable AI — not just a ranked list, but a reason you can check. Two papers are in journals. Eleven models and a dataset are on Hugging Face, including smaller GGUF builds that are easier to run.
I've shipped 20+ apps on iOS, Android, Flutter, and the web — digital signage, property tools, and client products. I also keep two Dart packages on pub.dev.
Soongsil University, Seoul
Completed February 2026 · CGPA 4.28/4.5
Field-Adaptive Dense Retrieval of Structured Documents
Muhammad Ali Jinnah University, Karachi
University of Sindh, Jamshoro
Journal of the Korea Multimedia Society
Vol. 28, No. 8, pp. 1001–1014, August 2025
How to search structured documents when each field needs its own treatment, so ranking stays fair across different layouts.
Applied Sciences
Vol. 14, No. 17, article 7881 · September 2024
Turns a question in plain language into a Cypher graph query, using BERT, GraphSAGE, and a transformer together.
Ecoda Co., Ltd., Seoul
Soongsil University, Seoul
Cublick Digital Co., Ltd., Seoul
Progatix, Karachi
Wiztech Pakistan (Pvt.) Ltd., Karachi
Document search, shop queries, and vision models — eleven models and a dataset
Document search
Finds similar documents even when the fields don't line up the same way.
View ModelSearch queries
Writes a search query that matches the kind of document you're looking at.
View ModelSummaries
Writes a short description that follows the document's own fields.
View ModelSmaller file
The smaller version of the description model.
View ModelQuery rewrite
Turns a messy shop question into a clear search query.
View ModelIntent
Figures out what the shopper is trying to do.
View ModelDataset
10,000 shop conversations used to train the rewrite models.
View DatasetVision-language
Reads a drawing and turns it into structured rooms and walls.
View ModelVision-language
Same vision reader, trained further so the geometry holds up better.
View ModelObject detection
Finds walls, doors, windows, and symbols in drawings.
View ModelSegmentation
Draws the actual shapes of those parts, not just boxes.
View ModelRetrieval (RAG), explainable AI, vision, speech — plus full-stack and mobile
PyTorch, TensorFlow, training, and getting models onto a machine
Training, prompts, and checking whether the answers hold up
Bi-encoders, RAG, and retrieval you can inspect
Why a result ranked, why a query was rewritten, what the model actually used
NumPy, Pandas, datasets, data visualization
Smaller GGUF files so models actually run on ordinary machines
Finding objects and outlining them in photos and drawings
YOLOv8 for boxes and masks in images and drawings
Models that look at images and text together
Workflows, webhooks, and glue between tools
Tying APIs together, including agents and search steps
Speech to text and audio transcripts
Spoken replies and cleaning up audio before it goes in
Swift, Objective-C, SwiftUI, UIKit, CoreData
Kotlin, Java, Jetpack Compose, Material Design
Cross-platform, Dart, State management, Animations
BLE, CoreBluetooth, Device communication
GPS, Google Maps, Apple Maps, Geofencing
Stripe, PayPal, Apple Pay, Google Pay
NestJS, Express, RESTful APIs, WebSocket
Laravel, CodeIgniter, WordPress, Symfony
TypeScript, RxJS, NgRx, Material Design
Components, Hooks, Redux, Context API
ES6+, jQuery, Bootstrap, responsive UI
Responsive design, Flexbox, Grid, Animations
Complex queries, optimization, stored procedures
NoSQL, Document stores, Aggregation pipelines
FAISS, Pinecone, Weaviate, Embeddings
Neo4j, Cypher queries, Knowledge graphs
Realtime DB, Firestore, Auth, Cloud Functions
Advanced SQL, JSON support, PostGIS
Version control, Collaboration, CI/CD workflows
Containerization, Docker Compose, Multi-stage builds
EC2, S3, Lambda, CloudFront, RDS
GitHub Actions, GitLab CI, Jenkins, Automation
Project management, Jira, Sprint planning
VS Code, Xcode, Android Studio, PyCharm
Retrieval work first, then the apps around it
Thesis work: retrieval that changes with the document's fields. The paper is out, and the models (plus smaller GGUF builds) are on Hugging Face.
Shop search that can follow a conversation. If someone says “the cheaper one”, it rewrites that into a real query.
Property management in one Flutter codebase for Android, iOS, and web. Rent, background checks, and video calls.
Live GPS tracking for security guards, including background running, battery care, and admin approval.
Digital signage with content picks, live stats, and remote control for many screens.
Emoji picker for Flutter — 2700+ emojis, 35+ categories, search. A few hundred downloads a week on pub.dev.
Small Dart helper for parsing and converting data. No extra packages. A hundred-plus downloads a week.
Procurement reporting in Flutter: tracking, approvals, and simple analytics.
School app for parents across more than one campus — grades, messages, and updates.
Native iOS/Android app talking over Bluetooth to Water Link Spin Touch devices, with PDF reports.
Also LolyHub, Rent College Pads, Auto Service Repair, Tap n' Save, HubPro Tracking, INDEL, and a few websites.
View All on GitHubIf you want to talk about retrieval, RAG, or putting a model into a product, write to me.