JUNREY
TEJAS

Architect · Engineer · Operator

Built the platform, then built the ISP that runs on it.

End-to-end ISP management platform for Exybit Network and Data Solution, an internet service provider I founded and operate in Cebu, and the business Exybit’s software solutions practice grew out of. Property-centric CRM wired directly into MikroTik routers over the RouterOS API: PPPoE provisioning, profile-swap suspensions, session kicks, prepaid billing with idempotent payments and automatic retry crons, a customer self-service portal, ticketing, and live CPE monitoring through a self-hosted UISP instance.

• Direct MikroTik RouterOS integration: subscriptions provision, suspend, and reactivate themselves on real hardware
• Fault-tolerant billing: idempotent payment handling with atomic gateway dedup; expiry crons re-attempt router changes when hardware is unreachable
• Customer portal with separate JWT auth domain, receipts, and self-service tickets
• Serving real paying subscribers in production

Engineer

AI-matched, referral-driven hiring for the US market.

Jobvious is a US referral-based hiring platform: an AI engine matches candidates to roles by semantic similarity, and a referral network lets people earn by bringing others in. It spans a candidate app (web plus native iOS and Android), an employer portal, an admin console, and real-time chat. I've built at the core of its business operations. Lumo, the AI support assistant I built and sold to Jobvious, now powers its in-product support chat.

• AI job matching on semantic embeddings and vector search: bi-directional candidate-to-role scoring, not keyword search
• Referral engine with device-fingerprint attribution linking referral clicks to app installs
• Two-way JobDiva (ATS) sync plus resume-parsing, background-check, and notification pipelines on queue-backed workers
• One codebase, many surfaces: React web, native iOS/Android via Capacitor, real-time chat over WebSockets

Engineer

One portal for a national operations company.

Unified business-operations portal for Lotus Commercial, Australia's leading commercial-kitchen services company, running on the order of 80,000 service jobs a year for thousands of clients. The portal houses integrations to multiple internal and external systems with realtime processing: unified job management across internal and external sources, invoicing, and operational dashboards in a single pane of glass.

• Unifies internal and external job pipelines into one operational view
• Realtime processing across multiple integrated systems
• Invoicing and dashboard tooling for a ~70-person national operation

Backend Engineer · 2022—2024

The rate-compliance engine behind Australian staffing.

Wage-rate and charge-rate compliance SaaS for the Australian market. It calculates employee pay rates and client charge rates that are both compliant and profitable, and generates quotes, schedules, agreements, and contracts. From 2022 to 2024 I contributed heavily to the backend API services at the core of the product, on infrastructure the company hosts itself.

• Two years of sustained contribution to the core backend API services
• Rate-calculation and document workflows: quotes, schedules, agreements, contracts
• Helped manage the company's own server environment alongside the product work

Creator

A software-development AI engine: API-first, CLI on top.

An AI engine for software development that can be consumed as an API or driven through its own client CLI. A NestJS service (my proprietary agentic engine, built on a frontier language model) with semantic caching (pgvector + fastembed), Redis-backed session persistence, and an MCP server for device management. The terminal client, @exybit/cli, is published on npm and talks to the engine over WebSocket.

• Semantic response caching with pgvector + fastembed embeddings
• Dual surface: REST/WebSocket API and a Model Context Protocol server
• @exybit/cli published on npm, a TUI chat client for the engine
• Powers AIDA (my self-extending personal assistant) and the chat assistant on this site, in production
• OAuth token refresh, hot-reloadable config, session persistence

Creator

Built it, ran it, sold it.

An AI support assistant for companies: an embeddable chat widget that answers customer questions from a business's own knowledge base. I built it, operated it as a product, and sold it: Lumo was acquired by Jobvious and now powers their in-product AI support chat.

• Embeddable support widget that answers from each business's own knowledge base
• Built, operated, and sold as a standalone product
• Acquired by Jobvious; now powers their in-product AI support chat

Creator

A self-hosted AI assistant that extends itself.

A personal and family AI assistant that runs entirely on my own infrastructure (driven from a Telegram family group and a mobile web app), handling expenses, savings goals, to-dos, reminders, notes, receipt capture, live web research, and on-demand financial analysis, all backed by my own Postgres and powered by my own Exybit AI engine so the data never leaves my stack. What sets it apart: when I ask for something it can't do yet, it interviews me, writes a spec, and an automated pipeline (code generation → CI gate → AI self-audit → auto-repair) opens a pull request I approve with one tap in Telegram. Then it deploys the new capability to itself. The assistant in the corner of this page is one face of it.

• Self-extending: a request becomes a spec, then a codegen → CI → self-audit → auto-repair pipeline that opens a PR I approve in one Telegram tap, and deploys to itself
• Self-auditing: logs how it is actually used and runs a proactive self-review, proposing concrete improvements instead of waiting to be told
• Two-brain design: a fast deterministic system of record (typed actions, schedulers, Postgres) paired with an on-demand agentic layer for research, analysis, and codegen, with each task routed to the right one
• Honesty by construction: guardrails stop it claiming work it did not do; failed builds route back with the real reason and a PR link