Hermes Agent Challenge

My Tech Stack

Next.js 14 + TypeScript, Tailwind CSS, Framer Motion, Clerk (auth), Supabase (Postgres), SSE streaming, Telegram Bot API, deployed via Docker + Coolify on DigitalOcean. AI: Hermes Agent, powered by MiniMax (I got their Pro plan free through a hackathon, which made running the agent possible).

How I Used Hermes Agent

Hermes is the engine behind every output. I leaned on its agentic capabilities:

• 
  Live web search for real competitor data — not hallucinated.


• 
  Streaming with live progress so you watch the agent think instead of a spinner.


• 
  Multi-step pipeline where the research is fed forward as context into the next three tasks for consistent positioning.


• 
  Structured output via a sentinel protocol, parsed into saved, typed results.


• 
  Human-in-the-loop approval between every step.

This was an agent problem, not a completion problem — it had to research the real world, produce several different artifacts, carry context between them, and report progress — all of which Hermes handles natively.

Future Improvements

• 
  Automated image generation — auto-generate logos, OG images, and landing-page visuals per run.


• 
  Dedicated backend — move the in-process job queue to a shared store (Redis/Supabase) for horizontal scaling.

I haven't built these yet simply because I don't have the capital for the extra infrastructure and image APIs right now. As a student (broken guy lmao) , I focused on shipping a solid core experience first.