Every project, written up in full.
Each project here has its own page with the architecture, the decisions behind it, what the numbers actually were, and where it falls short. No screenshots without an explanation, and no metric without the method that produced it.
- Projects
- 5, all documented
- Domains
- Agents · LLMs · Security · XR / Vision
- Code
- Public where the licence allows
- Write-ups
- Architecture, results, limits
The work
T-Rax — Multi-Agent Unity Code Generation
A LangGraph-style multi-agent pipeline — orchestrator, planner, and executor agents — that plans and generates C# Unity scene code end-to-end.
The problemBuilding one XR training scenario costs developer-weeks, so the organizations that need drills most end up with three of them instead of a library.
Read the case study →SmartReco
A four-stage DeFi security pipeline combining Slither static analysis with REVM-based dynamic transaction replay. Tested across 17 major protocols — including Aave V3, Compound V3, Uniswap V3, Balancer V2, and Lido — with zero vulnerabilities detected and 90%+ replay accuracy.
The problemStatic analyzers hand teams long lists of maybe — findings nobody can act on, and the alert fatigue that hides the real one.
Read the case study →VR Hand Interfaces on Quest 2
Replication of the CHI 2022 "Hand Interfaces" interaction technique across 11 virtual objects, migrated from the legacy Oculus Integration SDK to the Meta XR SDK, targeting Meta Quest 2.
The problemXR training content dies when the SDK it was built on is deprecated, stranding the headsets a school or a public body has already paid for.
Read the case study →Small-Model Fine-Tuning Ablations
Fine-tuned 1–3B parameter models (Phi-3-mini, Qwen2.5-1.5B, Gemma-2B) to study dataset-quality effects, using LoRA/PEFT and GGUF quantization across roughly 12–24 GPU hours.
The problemOrganizations whose data cannot leave the building are told their only option is a frontier model in someone else's cloud.
Read the case study →Gmail Assistant
An orchestrated n8n agent that drafts context-aware Gmail replies with Claude, negotiates and books meetings against live calendar availability, and writes a twice-daily priority-ranked inbox recap.
The problemRoutine correspondence eats staff hours — triage, scheduling back-and-forth, and no reliable picture of what is still pending.
Read the case study →More on the way
These are the projects that are finished and written up. Others are in progress — agentic RAG over private document sets, retrieval evaluation harnesses, and a few things from the day job that will need permission before they can appear here. The stack page lists the wider set of areas I work in; this page only lists what I can show you end to end.