Computer Science undergrad at NUST (Class of 2027) focused on Machine Learning,
Computer Vision, and production AI systems. I turn research ideas into working software —
from GPU-accelerated classical ML pipelines to clinical tools and agentic RAG platforms.
I've shipped real outcomes: a nuclei segmentation pipeline hitting 0.8802 Macro-F1
on PanNuke, a live clinical workflow product (Cepharix), computer vision work at
Infix Systems, and medical imaging research at MachVIS Lab.
My stack centers on Python, FastAPI, AWS, and modern ML tooling — with a bias toward measurable
results, clean engineering, and systems that hold up beyond demos.
Open to ML Engineering, Computer Vision, and Applied AI internships or full-time roles where
I can own hard problems end-to-end and deliver impact quickly.
• Contributed to early-stage development of a vehicle speed estimation and license plate recognition system.
• Annotated industrial safety data using Labelme.
• Built backend for a Windows virtual camera app, streaming real-time UDP video via FFmpeg as a low-latency system-level source for Zoom, Teams, and Meet.
• Learned the complete research workflow, including literature review and identifying research gaps.
• Executed research-driven data workflows including academic paper retrieval, exploratory data analysis, and structured documentation using spreadsheets for literature synthesis.
Built a production-grade fullstack trucking platform from scratch — featuring FMCSA-compliant ELD log generation, real-time route calculation via OSRM, HOS (Hours of Service) violation detection, interactive map with Leaflet, SVG log sheets, JWT authentication, trip saving, and a fully responsive glassmorphic UI. Deployed on Railway (Django) + Vercel (React) + Supabase (PostgreSQL).
Building an upcoming multi-tenant SaaS platform that helps researchers extract citation-grounded insights from scientific literature — turning dense PDFs into structured, trustworthy evidence for faster literature review.
Built a GPU-accelerated classical ML pipeline segmenting nuclei on 7,904 PanNuke images using 93 handcrafted features — with memmap streaming, Active Boundary Mining, GPU RFE (12× speedup), and an RF/XGBoost/LightGBM ensemble achieving 0.8802 Macro-F1.
Built a RAG system to answer academic policy queries from institutional documents — with ingestion, chunking, embedding, and retrieval pipeline for accurate context, integrated with LLMs to reduce hallucinations and ground responses.
Built a full clinical workflow prototype covering patient/case management, X-ray upload, AI landmark detection, interactive landmark refinement, and Ricketts cephalometric analysis with report generation — including drag-and-drop landmark editor with undo/redo.
@mananbyte
You can explore further by clicking on any of the profile cards above!