Building Digital
Solutions That Matter

Olist Query Studio

LLM • Graph Analytics

A local workbench that turns plain-English analytics questions about the Olist e-commerce dataset into Cypher, runs them against Neo4j, and currently matches 82 of 112 verified benchmark questions (73%). A local Ollama model writes structured queries from the graph schema; a validator, an EXPLAIN check, and a read-only database account keep every query safe, with one automatic correction attempt on failure.

Bug2JSON

LLM Fine-Tuning • Bug Triage

Turns free-text bug reports into schema-validated triage records: category, severity, priority, environment, reproducibility, probable cause, and next step. A FastAPI service and React reviewer UI compare a base Qwen model with a LoRA fine-tune. On a 30-case benchmark the untuned base model returns 100% schema-valid JSON with 0.34 category macro-F1, the baseline the fine-tune has to beat.

A full-stack web app that compares PDF/DOCX resumes with job descriptions or job links and returns ATS-style feedback, keyword gaps, and improvement suggestions. Built as a monorepo with a Next.js frontend, an Express/TypeScript API, shared Zod schemas, Supabase authentication, and analysis history, combining OpenAI-assisted feedback with deterministic scoring and heuristic fallbacks.

A local system monitoring dashboard. An Express backend polls CPU, RAM, GPU, storage, thermal, battery, process, network, and socket telemetry behind cached API endpoints, and a React/Vite UI renders live cards, time-series charts, process tables, and network and storage views, with confirmed process termination through the API.

Sort-Downloads

Python • CLI • Automation

A command-line tool I use daily to keep my Downloads folder tidy. It files every loose download into category, extension, and month folders and sends exact duplicates (same size and MD5) to the Recycle Bin, after a colour-coded preview of every move. Ported from my PowerShell script, with a parity test proving both versions file everything identically.