AI assistants & LLM integration

The systems behind it

• 2026 — Present

  Omnichannel AI Assistant

  One webhook, one app, every channel — WhatsApp, Messenger and Instagram

  An automated reply system serving every page and every WhatsApp number a company owns from a single deployment. The AI layer is abstracted behind a provider interface, so the model can be swapped without touching conversation logic — and the Meta app-review surface (privacy policy, terms, data deletion) ships with it.

  Role: Sole engineer — AI abstraction, channel routing and Meta review

  Node.jsExpressOpenAI SDKMeta Graph APIWhatsApp Cloud APIInstagram Messaging
• 2024 — Present

  Multi-Tenant WhatsApp Messaging Platform

  Campaigns, scheduling and AI replies across many client accounts

  A messaging platform serving multiple client accounts from one deployment: webhook ingestion, template management, scheduled jobs, an AI chat service, push notifications and invoice generation.

  Role: Backend engineer

  Node.jsExpressMongoDBAgendaFirebaseOneSignalSendGrid
• 2024 — 2025

  Real-Time Voice Translation

  Speech in, translated speech out — with the slow parts pushed off the request

  A Python speech service handles recognition and translation while a Laravel control plane manages speakers, subscriptions and quotas. Everything expensive runs through Redis-backed queues, so a slow model response never blocks a request — the same asynchronous pattern that carries the translation workload on the Bulum platform.

  Role: Full-stack — speech service, queueing model and admin control plane

  FastAPIOpenAISpeechRecognitionPyAudioNumPyLaravelRedis queues
• 2024 — 2025

  Bulum Platform

  Social platform with a queue-driven translation pipeline

  Real-time voice processing and multi-language support built on Redis-backed queues, so heavy translation work runs asynchronously instead of holding requests open.

  Role: Backend and infrastructure

  LaravelRedisQueue workersTranslation APIREST API