Webhook listener
Subscription verification, inbound event discovery, and signature validation against the original request body.
DEDICATED COMPONENTS / MESSAGING & AI
Useful building blocks.
Focused integrations.
Reusable across products.
Alongside complete products, I’ve established dedicated components for messaging and AI integration. Each addresses a specific external system while keeping the application’s own responsibilities clear.
MESSAGINGWhatsApp & Telegram
AIOllama & OpenAI
DESIGNFocused boundaries and explicit configuration
01 — WHATSAPP
Connect conversations
to the application.
Separate listener and bot libraries handle the two sides of WhatsApp integration: receiving verified webhook events and interacting with the Graph API.
Outbound messages
Text, interactive menus, and call-to-action links, with support for replies to provider message IDs.
Messages and media
Parsing text and interactive replies, recognizing media messages, and retrieving media metadata and downloads.
Host-independent contracts
Focused interfaces let an application own its HTTP endpoint and business behavior while reusing the WhatsApp integration.
02 — TELEGRAMBOT
Bot and client.
Distinct responsibilities.
TelegramBot brings together separate bot and Telegram user-client connectors, with supporting host, UI, and server projects.
Bot connector
Bot connection, message reception, chat-type handling, and document sending code.
User-client connector
Connection and verification workflows, chat discovery, message access, and file uploads through WTelegram.
Upload feedback
Upload handling includes progress models so a consuming application can track transferred data.
This work includes integration components and experimental application code; individual capabilities are described here without implying a complete hosted Telegram product.
03 — AI PROVIDERS
Choose where
inference runs.
Zilkode.AI.Ollama and Zilkode.AI.OpenAI establish separate integration foundations for local and hosted AI. The idea is interchangeability: keep the application’s AI work separate from the system that executes it.
Zilkode.AI.Ollama
Local-provider endpoint definitions for chat, embeddings, and model discovery, with an explicit model identity type.
Zilkode.AI.OpenAI
Hosted-provider endpoint definitions for Responses and embeddings, including embedding-dimension support checks for compatible models.
Application AI workConfigured provider
Ollama · localOpenAI · hosted
Reduce API costs with local execution
Running suitable workloads locally through Ollama can reduce paid API usage. Hosted OpenAI remains an option when its models better fit the task. Local execution still carries hardware, power, and maintenance costs.
Keep the choice explicit
Provider choice belongs at the integration boundary. The goal is to change where AI work runs without scattering provider-specific behavior throughout the product.
Respect model compatibility
Interchangeability requires attention to model capabilities and embedding compatibility. Switching an embedding model can require regenerating stored vectors.
The diagram shows the integration concept. The inspected Zilkode projects contain endpoint and shared HTTP foundations; it does not claim that those libraries alone implement a complete provider-switching runtime.
SEE THE PRODUCTS
Selected work