Projects & Architectures

OpenVett is an enterprise AI-powered hiring and technical assessment platform for generating, delivering, scoring, and managing candidate evaluations. Key Capabilities: • AI assessment generation: Creates customized, role-specific tests from job descriptions and skill requirements across Technical, Reasoning, and Problem-Solving tracks. • Timed candidate exam portal: Candidate-facing secure test sessions with timers, progress tracking, and fraud prevention controls. • Automated AI evaluation: Intelligent grading with skill breakdowns, strength/weakness matrices, and hiring recommendations. • Admin dashboard: End-to-end management of jobs, candidate pipelines, question banks, and audit logging. • Infrastructure & Ingress: Self-hosted Docker Compose stack with automatic GPU/CPU Ollama model selection (Gemma3), PostgreSQL + pgvector, Redis + ARQ workers, and Cloudflare Tunnel prod ingress.

Full-stack coding-activity tracking platform that ingests heartbeat telemetry from code editors, computes real-time daily programming metrics, and surfaces high-density analytics dashboards. Key Capabilities: • Custom VS Code Extension: Built a dedicated VS Code VSIX extension ingesting typing and save events with debounced batching. • Zed Editor LSP Bridge: Engineered a custom stdio Language Server Protocol (codeowl-ls) binary bridging Zed editor sessions to the CodeOwl ingestion server. • Edge Runtime Hardening: Designed edge-safe session verification inspecting auth cookies directly to prevent TCP socket lockups in Next.js 15 App Router middleware. • Resilient Ingestion: Rate-limited API ingestion endpoint with daily and weekly metrics rollups, language distribution charts, and project timelines.

Production backend engineered for a multi-tenant fashion e-commerce platform with clean architecture and strict type safety. Key Capabilities: • Modular Architecture: Built using NestJS 10, TypeScript, Prisma, and PostgreSQL with full multi-role JWT authentication (Admin, Boutique Vendor, Customer). • Inventory & Stock Restoration: Atomic inventory management restoring variant stock and parent aggregate levels on order cancellations. • Flutterwave Payments: Idempotent payment handling with webhook signature verification, transaction auditing, and double-spend protection. • Asynchronous Workers: BullMQ and Redis queues handling order expiration, abandoned checkout cleanup, and transactional notifications via Nodemailer. • Deployment: Production Docker Compose on VPS with health checks, seed compilation, and automated deployment script.

Autonomous zero-cost market research agent SaaS running completely on local AI models without third-party API dependencies. Key Capabilities: • Autonomous Agent Pipeline: Plan → Search (SearXNG) → Read (Readability) → Ingest & Embed (pgvector + nomic-embed-text) → Synthesize Report. • Data Synthesis: Algorithmic competitor clustering, pricing benchmarks, TAM/SAM/SOM financial sanity validation, and weighted go/no-go scorecards. • Monetization & Quotas: Lemon Squeezy subscription billing with append-only usage ledgers, monthly quota resets, and Better Auth authentication. • Local Inference: Zero-cost operation powered by host Ollama running Gemma 3 models, eliminating external API usage costs.

Suite of privacy-first, on-device mobile applications published under the SquareAtom indie studio brand. Key Capabilities: • Finbit: Private on-device expense tracker built with Flutter, Riverpod, and Drift SQLite with encrypted PIN lock and local CSV/JSON backup. • Momentum: Offline-first habit and mood tracker featuring "Adaptive Habit Rescue" to preserve streaks dynamically. • Moonpage: Distraction-free, local-first EPUB/PDF reader built with Expo featuring full offline annotations and reading analytics. • CI/CD Pipeline: Self-hosted Dockerized GitHub Actions runner automatically building, signing, and uploading releases directly to the Google Play Console internal track.

High-throughput asynchronous job queue and concurrency backend built in Go, accompanied by Careerlens AI job application evaluator. Key Capabilities: • Worker Pool Concurrency: Idiomatic CSP pattern using buffered channel backpressure, dynamic worker goroutines, and graceful shutdown. • Storage Engine: Dual-repository architecture supporting SQLite in WAL mode with connection limits and thread-safe memory storage with sync.RWMutex deep cloning. • Standard Library Routing: Engineered with Go 1.22+ ServeMux ("METHOD /path/{id}") and structured JSON logging via slog. • Careerlens Streaming: AI evaluator service featuring Server-Sent Events (SSE) streaming, SSRF-guarded job scraping, and docx/pdf analysis.

Complete machine learning classification pipeline to predict customer churn using the Telco Customer Churn Dataset. Implements exploratory data cleaning, outlier handling, feature scaling, model training, hyperparameter tuning, and evaluation using Support Vector Machine (SVM) classifiers.

Full-featured learning management platform designed to streamline course delivery, student tracking, and instructor content management in an intuitive digital classroom environment.