I am a software engineer who likes building products end to end — from mobile apps and backend systems to AI tooling and experimental ML systems.
Right now, I am building Infimium-agent, a local-first context layer for AI coding agents. The goal is to help coding agents understand a project beyond a single prompt by giving them persistent context, semantic code search, repository awareness, and better continuity across sessions. I am also working on something deeper around how agent-driven software changes are planned, traced, and verified. That part is still private for now.
Alongside Infimium, I am building KlubEATS, a Flutter-based platform for enterprise food ordering and vendor operations. It brings employee ordering, meal booking, QR-based entitlements, payments, vendor workflows, feedback, analytics, and repeat-ordering systems into one product.
I am also working on Playground.ai, a visual canvas for designing app and web interfaces and turning them into usable UI code in your preferred stack - React, CSS, Dart, and more. The idea is to move from visual exploration to actual components and pages without breaking the flow between design and development. It is currently under construction and releasing soon.
Outside product work, I spend time experimenting with applied machine learning and model architecture. I built a LightGBM-based fraud detection system over 6.36M+ transactions, using SMOTE for class imbalance and SHAP to understand the strongest fraud signals Notebook. I also built an LSTM-based stock-price prediction pipeline in Python with automated market-data collection, preprocessing, sequence generation, and evaluation using RMSE and MAE Notebook.
More recently, I have been researching Salience-Gated Slot Memory for State-Space Models (SG-SSM) — a bounded, content-addressable memory mechanism that runs alongside a standard SSM recurrence. The idea is to preserve rare or high-salience information in dedicated memory slots instead of forcing everything into a shared low-rank state, then selectively combine that memory with the normal SSM output at read time. Research notebook
Most of the things I build sit somewhere between product engineering, AI systems, developer tooling, and applied research. I like working close to the implementation, understanding why a system behaves the way it does, and turning rough ideas into things people can actually use.
Mobile
Flutter · Dart · Firebase
AI / ML
Python · LightGBM · LSTM · SHAP · SMOTE · Embeddings · Semantic Search · FastEmbed · Tree-sitter · Claude API
Agent Systems
MCP · Context Engineering · Local-first AI · Codebase Indexing · Retrieval Pipelines
Web & Backend
TypeScript · JavaScript · React · Next.js · Node.js · Express · PostgreSQL · SQLite
Cloud & Infra
AWS · GCP · Docker · Kubernetes · Git · GitHub