I'm a data scientist and software engineer who turns messy data and hard problems into systems that actually run. Most of my depth is in data science and machine learning — exploratory analysis, statistics, predictive modeling, and computer vision — paired with the full-stack engineering it takes to move a model from a notebook into production.
I like understanding the layer below the one I'm working on, whether that's the math behind a model or the service that serves it. Depending on what a problem needs, I move between data wrangling and analysis in Python, PyTorch training loops, and the APIs and backends that deliver the results.
Data science and machine learning are where I go deep. Software engineering is the throughline that keeps the work reproducible, and full-stack thinking is how I make it usable.
Open to · Data Science roles · AI / ML Engineering roles · Software Engineering roles · Open-source collaboration
Languages
Frontend
Backend & Databases
AI / ML & Computer Vision
Cloud, DevOps & Tooling
| Domain | Proficiency | Focus Areas |
|---|---|---|
| Computer Vision |  |
MediaPipe face mesh, YOLOv8 detection, OpenCV, real-time webcam inference |
| Deep Learning | |
PyTorch, LSTMs, CNNs, time-series forecasting |
| Multimodal AI / LLMs |  |
CLIP embeddings, vision LLMs (Llama 4 Scout via Groq), RAG patterns |
| NLP & Embeddings | |
Sentence-BERT, spaCy, semantic search |
| Classical ML & Data Science | |
scikit-learn, predictive modeling, EDA, dashboards |
| MLOps & Serving | |
ONNX, Docker, FastAPI model serving, Qdrant vector DB |
Bisondb — A Document Database Built From Scratch (C++20)
A document-oriented database written in C++20 with no third-party storage or networking libraries. It implements a BSON storage engine, append-only collections, hand-written on-disk B+Tree indexes, a query engine with explain plans, a TCP server (bisond), and an interactive shell (bisonsh).
| Aspect | Detail |
|---|---|
| Stack | C++20, custom BSON, on-disk B+Tree, TCP sockets, CLI shell |
| Scale | Append-only collections with persistent indexing |
| Performance | Hand-written B+Tree indexes and query explain plans |
| Security | Zero external storage / networking dependencies, full control of the stack |
| Impact | End-to-end database internals: storage, indexing, query, networking, shell |
| Repository | Bisondb · Project Site |
This is the project I point to when someone asks whether I understand what a database does under the hood. Every layer, from how a document hits the disk to how an index lookup resolves, is something I wrote and can explain.
Multimodal Product Intelligence — Vision LLM + Vector Search API
A FastAPI backend for AI-driven product analysis. It pairs Groq Vision (Llama 4 Scout) for image understanding with CLIP embeddings and Qdrant for semantic retrieval, exposed through a typed API and a separate frontend.
| Aspect | Detail |
|---|---|
| Stack | FastAPI, Groq Vision (Llama 4 Scout), CLIP, Qdrant, Python |
| Scale | Vector search across product image and text embeddings |
| Performance | CLIP embeddings with Qdrant approximate-nearest-neighbour retrieval |
| Security | API-key authentication and typed request / response schemas |
| Impact | Combined image and text product understanding with semantic search |
| Repository | API · Frontend |
A practical look at how vision models, embeddings, and a vector database fit together into one service rather than three disconnected experiments.
Real-Time Driver Drowsiness & Distraction Detection — Computer Vision
A live driver-monitoring system that fuses three signals: MediaPipe face mesh for eye and head state, YOLOv8 for phone detection, and a PyTorch LSTM that reads the temporal pattern of those landmarks. A Streamlit UI runs it against a live webcam.
| Aspect | Detail |
|---|---|
| Stack | MediaPipe, YOLOv8, PyTorch LSTM, OpenCV, Streamlit |
| Scale | Live webcam, frame-by-frame inference |
| Performance | Temporal LSTM over facial landmarks for drowsiness classification |
| Security | Runs locally; video never leaves the device |
| Impact | Driver-safety monitoring through multi-signal sensor fusion |
| Repository | Driver Drowsiness Detection |
Three models doing different jobs, joined into one real-time decision instead of a single classifier guessing at everything.
TaskForge — Multi-Tenant SaaS Project Management Platform
A production-grade, multi-tenant SaaS platform built on Django and DRF. It covers the parts a real product needs: JWT auth, role-based access, Stripe billing, async notifications via Celery, real-time task updates over Channels, and a Dockerized CI/CD pipeline with a real test suite.
| Aspect | Detail |
|---|---|
| Stack | Django, DRF, PostgreSQL, Redis, Celery, Channels |
| Scale | Multi-tenant with role-based access control |
| Performance | Background work via Celery, real-time updates via Channels / WebSockets |
| Security | JWT auth, RBAC, and production-grade testing |
| Impact | Stripe billing, async notifications, Dockerized CI/CD deployment |
| Repository | Backend · Frontend · Desktop |
The architecture is the point here: tenancy, billing, async jobs, and real-time messaging are the pieces that separate a demo from something you could actually deploy.
Face Verification System — Cross-Platform Desktop CV
A face-verification system delivered as a native desktop application using Tauri 2, with a companion web frontend. Verification runs on-device.
| Aspect | Detail |
|---|---|
| Stack | Tauri 2, computer vision, web frontend |
| Scale | Cross-platform native desktop build |
| Performance | On-device face verification |
| Security | Local biometric processing, no cloud round-trip |
| Impact | Identity verification packaged as an installable desktop app |
| Repository | Desktop · Frontend |
Taking a vision model from a notebook into a shippable desktop app, where startup time, packaging, and on-device performance all suddenly matter.
ArenaBinAllocator — A Custom Memory Allocator in C
A memory allocator written in C that manages a preallocated arena. It reimplements malloc, free, and realloc, using a binning strategy for small allocations and block merging to keep fragmentation down.
| Aspect | Detail |
|---|---|
| Stack | C, manual memory management |
| Scale | Preallocated arena with size-class bins |
| Performance | Binning for fast small allocations, block merging to reduce fragmentation |
| Security | Explicit, bounds-aware allocation logic |
| Impact | malloc / free / realloc rebuilt from first principles |
| Repository | ArenaBinAllocator |
The kind of project that changes how you read every line of C afterwards, because you've seen what the allocator is doing underneath.
More on GitHub · Battery Degradation Forecasting (LSTM) · SpaceX Falcon 9 Landing Prediction · Credit Card Fraud Detection · Hospital Readmission Prediction · Electronics Store (Node / Express / MongoDB) · Furniture Image Classification
2023 — Present
Designing and shipping end-to-end systems across AI, computer vision, and full-stack development.
- Built a document database engine in C++20 with custom storage, on-disk B+Tree indexing, a query engine, and a networked server.
- Developed real-time computer-vision pipelines fusing MediaPipe, YOLOv8, and PyTorch for live inference.
- Architected a multi-tenant SaaS platform with billing, async processing, real-time updates, and CI/CD.
- Shipped a multimodal AI API combining vision LLMs, CLIP embeddings, and Qdrant vector search.
C++ Python PyTorch Computer Vision Django FastAPI Docker System Design
| Recognition | Details |
|---|---|
| Database engine from scratch | Designed Bisondb with a custom BSON store and hand-written B+Tree indexes in C++20 |
| Multimodal AI pipeline | Shipped a product-intelligence API combining vision LLMs, CLIP, and Qdrant |
| Production SaaS architecture | Built TaskForge as a multi-tenant platform with Stripe billing and real-time updates |
| Systems programming | Implemented a custom arena memory allocator in C with binning and block merging |
Data Science Specialization — Coursera
A multi-course specialization covering the end-to-end data-science workflow — data wrangling and exploratory analysis in Python, applied statistics and probability, machine-learning modeling and evaluation, and communicating results through clear visualization and reporting.
Abdullah Masood:
learning:
- Distributed systems and consensus
- MLOps and model serving at scale
- Rust for systems programming
building:
- Bisondb # document database in C++20
- Multimodal product intelligence API
exploring:
- Vector databases and semantic search
- Real-time computer-vision pipelines
- LLM application architecture
ask_me_about:
- Database internals
- Computer vision
- Full-stack SaaS engineering
open_to:
- Software Engineering roles
- AI / ML Engineering roles
- Open-source collaboration
From B+Trees to UIs, I like building the whole stack.