In April 2023 I joined the Antler accelerator in Ho Chi Minh City and co-founded Synna as CTO. The thesis: companies sit on massive internal knowledge (documents, emails, Slack, calendars) with no way to search across it intelligently. This was months after ChatGPT launched — AI engineering barely existed. No tool calls, no agent frameworks, no workflow builders. We were building what would later be called agentic workflows, from scratch.

I owned the backend end to end. Users could register, create organizations, connect data sources (Gmail, Slack), and ask questions across all of them in natural language, updating in real time as new documents arrived. The architecture was designed around iteration speed and security: I split indexing into a network-isolated component to minimize attack surface and encrypted all stored credentials. Stack: Python on GCP Cloud Run, Terraform, CI/CD, OpenAI for embeddings, Pinecone for vectors. The working prototype came together in about three and a half weeks.

In an accelerator, the MVP is the argument. Pitch decks are hypothetical; a working product is not. The prototype showed the approach was sound, the integrations worked, and the architecture could scale — the primary factor in securing $120K in pre-seed funding.