AI Resources

Research, experiments, and practical notes on AI agents, tools, models, and emerging workflows.

This repository is where I document what I am learning while building agentic systems, evaluating AI-enabled product workflows, and experimenting with Hermes Agent.

What's inside

Hermes Research

Recurring research reports on:

• New AI tools and agent workflows
• GitHub projects with practical product potential
• Local-first and BYOK creator tools
• Agent skills, orchestration, evaluation, and security
• Product remix ideas based on emerging open-source projects

AI Business Workflows

A collection of practical AI and automation workflow ideas organized by publication date.

Learning

Hands-on learning material covering prompt contracts, structured outputs, acceptance criteria, failure modes, and production-minded AI product design.

Insight2Spec API

A minimal FastAPI service that turns unstructured user feedback into structured product insights, including themes, opportunities, and experiments.

The API supports two modes:

• mock — deterministic local output with no LLM calls
• openrouter — live structured analysis through OpenRouter

Run Insight2Spec locally

make bootstrap
make smoke
PYTHONPATH=. .venv/bin/uvicorn app.main:app --reload

Environment variables

• INSIGHT2SPEC_ANALYZE_MODE — mock or openrouter (default: mock)
• OPENROUTER_API_KEY — required for openrouter mode
• OPENROUTER_MODEL — optional model override
• OPENROUTER_TIMEOUT_SECONDS — optional request timeout

Quick API check

curl -s http://127.0.0.1:8000/health | jq

curl -s -X POST http://127.0.0.1:8000/analyze \
  -H 'Content-Type: application/json' \
  -d '{
    "feedback": [
      "Search is slow when catalog size grows",
      "Users want saved filters for repeat workflows"
    ]
  }' | jq

All API failures return a consistent machine-readable error contract. Run make smoke before committing changes.

Research cadence

The Hermes research reports are published twice each week, on Tuesday and Friday mornings.