Skip to content
View skiiyuru's full-sized avatar

Block or report skiiyuru

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
skiiyuru/README.md

Stephen Kiiyuru

LinkedIn

Current

Pursuing an MSc in Artificial Intelligence Systems at the University of Trento.

My first year has been hands-on across several areas of modern AI:

  • Conditional image retrieval with CLIP: combining reference images with text attributes in embedding space, using geodesic composition and LoRA adaptation on CLIP's text encoder.
  • Language modeling from scratch: training a GPT-2-style model on Penn TreeBank and fine-tuning GPT-2-small with a from-scratch LoRA implementation, plus multi-task intent classification and slot filling on ATIS with GPT-2 and BERT.
  • 3D human reconstruction: currently training and benchmarking a CVPR 2025 Gaussian Splatting method (RoGSplat) on a high-resolution 4D human dataset (VolHuMe), including compiling CUDA extensions for Blackwell architecture and evaluating whether template-anchored approaches solve problems that standard NeRF/3DGS methods face on close-range captures.
  • Multi-agent systems: a BDI delivery agent paired with an LLM-powered agent that interprets natural-language missions and coordinates through tool-calling and socket-based delegation.

I'm using these projects to explore where my interests converge most meaningfully — vision, language, 3D, and interactive systems. My engineering background gives me a product and delivery perspective; my Master's studies are building the ML foundations underneath. I'm keeping my long-term direction open while I identify the problems where I can contribute most effectively.

Background:

Co-Founder of The Discovery Centre | Software Engineer

Organization

Motivation

Empowering children with coding and science skill.

Key Projects

Imla

AI powered transcription tool tailored to bilingual dialects in Kenya. Take a look

The Discovery Centre

Facilitating science and coding programs in primary and secondary schools. Take a look

Leadership & Community Impact

Executive Leadership at The Discovery Centre

  • Transformed traditional educational models through practical and engaging experiences.
  • Built and maintained strong relationships with key stakeholders from partner schools.
  • Organized annual free 3-week holiday programs for children aged 6 and above in collaboration with Goethe Institute Kenya.

Industry Engagement

  • Advocates for early adoption of software development skills among children.
  • Volunteers as a coding instructor, teaching HTML, CSS and Python to primary school children.
  • Shares insights on tech through portfolio blog.

Connect With Me

Pinned Loading

  1. kiiyurus.space kiiyurus.space Public

    My portfolio website

    XSLT

  2. compositional-image-retrieval compositional-image-retrieval Public

    A system that, given a reference image and a set of positive (`+`) and negative (`−`) text constraints, retrieves images that preserve the reference's identity while satisfying the constraints.

    Jupyter Notebook

  3. NPCurious NPCurious Public

    Implementation of a BDI agent that competes in the online game, DeliverooJS

    JavaScript