Local-first AI news digest experiments by Ameen Demiry.
AI Digest explores one product idea through four proof-of-concept tracks: turn noisy AI news, videos, benchmarks, and research signals into a structured daily briefing with source provenance, validation, diagnostics, and a polished HTML archive.
| POC | Folder | What it proves | Status |
|---|---|---|---|
| 1. LLM Pipeline | llm_pipeline/ |
Deterministic ingest, enrichment, validation, and report rendering. | Stable baseline |
| 2. Hermes Multi-Agent | agentic/hermes/ |
Four-role kanban crew: Concierge, Researcher, Librarian, Synthesizer. | Reference implementation |
| 3. Kaggle AI Agents | agentic/kaggle_ai_agents/ |
Course/capstone sandbox for skills, tool use, security gates, and evaluation. | Standalone POC |
| 4. Single Agent + Skills | agentic/single_hermes_agent/ |
Target direction: one host agent with progressively loaded skills. | Design stub |
- Auditable AI output: every digest story carries source metadata and passes deterministic grounding checks before publishing.
- Local-first execution: the main experiments are designed around local Python, Ollama-compatible LLM calls, and committed fixtures where possible.
- Multiple agent architectures: the repo compares a staged batch pipeline, a multi-agent kanban crew, and a simpler single-agent-with-skills direction.
- Portfolio-grade reporting: generated HTML reports include story cards, categories, provenance, leaderboards, archive pages, and diagnostics.
app/ is the approved website payload for mameen.github.io/AI_Digest. Treat those reports and diagnostics as canonical published artifacts.
| Artifact | Path |
|---|---|
| Website index | app/index.html |
| Approved report index | app/reports/index.json |
| Approved diagnostics index | app/diagnostics/index.json |
| Latest approved report | app/reports/20260728120000.html |
| Latest approved diagnostics | app/diagnostics/20260728120000.diagnostics.html |
Source POC folders may contain development outputs. When archiving, prefer app/ as the truth for what was approved and published.
# Bootstrap dependencies
python admin/manage.py bootstrap
# Run the stable pipeline baseline
python run.py --start 2026-07-29 --history 10
# Run tests
python run_tests.pyHermes requires a local Hermes/Ollama setup:
python agentic/hermes/admin/manage.py bootstrap
python agentic/hermes/admin/manage.py go --start 2026-07-09 --history 10 --freshKaggle POC commands live in agentic/kaggle_ai_agents/HOWTO.md.
SKILLS/ contains repo-assistant skills for contributors and coding agents. These are not product/runtime skills. Runtime skill experiments live under agentic/, especially agentic/single_hermes_agent/.
| Topic | Read |
|---|---|
| Documentation index | docs/README.md |
| Contributor/agent rules | AGENTS.md |
| Agentic POCs | agentic/README.md |
| Shared library notes | lib/README.md |
| Pipeline POC | llm_pipeline/README.md |
| Hermes POC | agentic/hermes/README.md |
| Kaggle POC | agentic/kaggle_ai_agents/README.md |
| Single-agent design | agentic/single_hermes_agent/docs/ideation.md |
Author: Ameen Demiry Portfolio: demiry.net GitHub: mameen/AI_Digest
AI Digest is released under the MIT License. Third-party dependencies retain their own licenses; see the dependency files and bundled notices where applicable.