NN Landscape Monitor
shehral/nn-landscape
Dashboard de inteligência AI/ML atualizado continuamente por agents
Ver no GitHubDashboard inteligente de monitoramento do panorama AI/ML que agrega e classifica notícias, papers e artefatos da web aberta, filtrados por relevância estratégica. Construído por agents Claude que rodam automaticamente a cada 6 horas, ingerindo dados de arXiv, Hacker News, RSS curados e GitHub Trending, com scoring por IA e geração de implicações editoriais.
Instalação
npx convez-ai shehral/nn-landscape1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Monitoramento contínuo de tendências em AI/ML relevantes ao negócio
- Identificação automática de papers e notícias com impacto estratégico
- Agregação editorial de múltiplas fontes com scoring por critérios customizados
Veredito

Projeto sofisticado que demonstra uso avançado de agents Claude em pipelines de dados reais com cron automation. Combina ingestão de múltiplas fontes, deduplicação, scoring LLM e renderização em um ciclo completamente automatizado — caso de uso maduro para times que precisam de inteligência competitiva contínua sobre AI/ML.
README
A tradução para o português deste README ainda está na fila.
Nanonets AI Landscape Monitor
A continuously-updated, public-shareable dashboard that aggregates and ranks AI/ML news, papers, blogs, and other artifacts from the open web, filtered and scored for strategic relevance to Nanonets.
The dashboard is built by Claude Code agents on a 6-hour cron. Every six hours a scheduled remote agent ingests items from arXiv, Hacker News, curated RSS, and GitHub trending; scores them on four axes (Document AI, Competitive, Frontier, VLM research); writes a 3-line "Implication for product / research / action" framing for the top items; surfaces emerging patterns and AI-partner questions; and commits the regenerated HTML. GitHub Pages serves the result.
Architecture
/schedule cron (every 6h) → remote Claude Code agent
→ reads .claude/skills/build-edition/SKILL.md (the playbook)
→ python -m landscape.cli ingest (HTTP fetch, no LLM)
→ python -m landscape.cli dedup (canonical URL + Jaccard fallback)
→ agent: score (Haiku subagent) (LLM judgment)
→ agent: frame top-N + trend pass + AI-partner pass (LLM judgment)
→ python -m landscape.cli render (Jinja2 → HTML)
→ python -m landscape.cli publish (validate, commit, push)
→ GitHub Pages serves docs/
The agent is the editorial mind. Python at the edges. Pydantic models validate every phase boundary.
Local development
Requires Python ≥ 3.11 and uv.
uv sync
python -m landscape.cli build --stub
open docs/index.html
The --stub flag uses deterministic synthetic scoring so the renderer can
be exercised without burning agent tokens.
Re-tuning relevance
The agent's notion of "relevant to Nanonets" lives entirely in
data/nanonets_context.md. To shift what gets
surfaced or how it's framed, edit that file and commit. The next build
will use the new context.
Adding sources
Edit data/sources.yaml. For new RSS feeds, just add
to the list. For new source types (e.g., Bluesky), add a new class in
src/landscape/ingest.py implementing the Source protocol.
Layout
nn-landscape/
├── .claude/skills/build-edition/ # agent playbook + editorial guide
├── data/ # context bundle, sources, weights
├── src/landscape/ # the package
├── templates/ # Jinja2 templates
├── docs/ # GitHub Pages root (rendered HTML)
├── state/ # seen.json + questions_for_team.md
└── tests/ # unit + integration
Documentation
docs/about.html— public-facing: coverage gaps, attribution policy, disclaimers, build provenance.docs/brainstorms/— design history.docs/plans/— implementation plans.state/questions_for_team.md— the AI agent's accumulating questions for the team.
Status
v1 in active development. Indexing is currently disabled
(docs/robots.txt: Disallow: /) until the editorial format is reviewed
and approved.
License
Internal Nanonets project. License TBD on public release.