Name: AI News Cafe Tagline: An AI-powered journalism chatbot for B2B media professionals, combining Claude AI with RAG-based knowledge retrieval and real-time web search to deliver credible, ethically-grounded media intelligence. Current status: Live First commit / project start: Late 2025 / Early January 2026 (v2.0.0 released January 8, 2026)
AI News Cafe is a WordPress-based AI journalism platform built around an intelligent chatbot for B2B media professionals. The chatbot, powered by Anthropic’s Claude API, integrates with a Pinecone vector knowledge base (via AI Engine), Tavily real-time web search, and a comprehensive system prompt focused on journalism ethics, AI best practices, and media industry expertise. The platform is deployed on a WordPress site running the Newspaper theme, with the chatbot embeddable via shortcode across any page.
AI News Cafe is purpose-built for the B2B media industry — not a generic chatbot repurposed for journalism. Its default system prompt encodes media management consulting expertise, journalism ethics guardrails, AI best-practices frameworks, source credibility lists (100+ approved domains), and source exclusion lists (content farms, low-credibility sites). The guardrails system ensures factual accuracy, prevents hallucination, and maintains editorial standards that generic AI chatbots do not enforce.
Platform: WordPress plugin (PHP) on a Newspaper-themed WordPress site Deployment: Self-hosted WordPress with external API dependencies (Anthropic Claude, OpenAI Embeddings, Pinecone, Tavily) Primary interface: Embeddable chat widget via [ainc_chat] shortcode; WordPress admin settings panel with tabbed configuration
Primary user: B2B media professionals — editors, publishers, media managers, newsroom technologists Secondary users: Journalism educators, media consultants, content strategists in publishing organizations User environment: Embedded on the AI News Cafe website (ainewscafe.com); accessible from any page where the shortcode is placed
B2B media professionals need reliable, context-aware AI assistance that understands their industry — newsroom workflows, audience development, content strategy, subscription models, and emerging AI journalism practices. Generic AI chatbots (ChatGPT, Gemini) lack domain-specific knowledge bases, journalism ethics guardrails, and credibility-filtered source lists, making them unreliable for professional media decision-making.
| Need | How the product addresses it | Source of evidence |
|---|---|---|
| Industry-specific AI guidance for media professionals | System prompt with deep media management consulting expertise, 2026 industry context, and authoritative source lists | ai-news-cafe-chatbot/README.md system prompt section |
| Ethical AI journalism guardrails | Comprehensive guardrails system with accuracy requirements, source credibility filtering, and editorial standards | ai-news-cafe-chatbot/GUARDRAILS.md, GUARDRAILS-ADDITION-SUMMARY.md |
| Knowledge-augmented responses with real-time information | RAG via Pinecone vector DB + Tavily web search for current news and research | ai-news-cafe-chatbot/includes/class-ainc-pinecone-api.php, class-ainc-tavily-api.php |
Media professionals relied on generic AI chatbots (ChatGPT, Copilot) that lack journalism-specific guardrails, credibility filtering, and industry knowledge bases. They supplemented with manual research across multiple sources, industry publications, and consulting engagements — a fragmented and time-consuming process.
Primary category: AI-powered journalism / media intelligence tools Market maturity: Emerging (AI journalism tools are nascent; most newsrooms are experimenting, not deploying purpose-built tools) Key dynamics: Newsrooms are under pressure to adopt AI while maintaining editorial standards. The AP, Reuters, and major publishers have established AI guidelines, but few tools exist that embed those standards into the AI interaction layer. ⚡
| Product / Company | Approach | Strengths | Key gap this project addresses | Source |
|---|---|---|---|---|
| ChatGPT / OpenAI ⚡ | General-purpose AI assistant | Broad knowledge, large context window | No journalism ethics guardrails, no credibility-filtered sources, no media industry knowledge base | ⚡ Public product |
| Google Gemini ⚡ | General-purpose AI with search integration | Real-time search grounding | No editorial standards enforcement, no B2B media specialization | ⚡ Public product |
| AI Engine (WordPress) ⚡ | WordPress AI chatbot with Pinecone RAG | Good WordPress integration | Generic chatbot framework; requires extensive customization for journalism use case | ⚡ Public plugin |
AI News Cafe positions as the first AI chatbot purpose-built for B2B media professionals, combining domain-specific expertise with journalism ethics guardrails that generic AI tools cannot provide. It demonstrates that responsible AI journalism requires not just better models, but better knowledge curation, source credibility controls, and editorial-standards enforcement at the system prompt level.
The product’s defensibility rests on its curated journalism knowledge base (Pinecone-indexed), its comprehensive system prompt encoding media consulting expertise and ethical AI guidelines, its source credibility and exclusion lists (100+ approved domains, explicit exclusion of content farms), and its guardrails architecture that would require significant domain expertise to replicate.
Requirements were developed iteratively through Cursor AI development sessions. The initial plugin build document (AI News Cafe Knowledgebase/AI News Cafe Plugin Build.md) defined the core architecture, and v2.0.0 requirements were captured across multiple focused update documents (database fix, admin settings fix, UI enhancements, guardrails addition).
Hard constraints:
Explicit non-goals:
| Decision | Alternatives considered | Rationale | Evidence source |
|---|---|---|---|
| Claude as primary LLM (not GPT-4 or Gemini) | OpenAI GPT-4, Google Gemini | Claude’s instruction-following and safety alignment suited journalism guardrails; cost-effective for long system prompts | ai-news-cafe-chatbot/README.md, model selection in admin |
| Pinecone via AI Engine plugin (v2.0) | Direct Pinecone integration only | AI Engine provides UI for knowledge base management; reduced custom code for RAG pipeline | CHANGELOG-v2.0.0.md — removed AI Power, unified on AI Engine |
| Removal of AI Power integration in v2.0 | Keep dual AI Power + AI Engine support | AI Power caused conflicts and complexity; single integration path improved stability | CHANGELOG-v2.0.0.md, PLUGIN-UPDATE-COMPLETE.md |
| Comprehensive source credibility lists in system prompt | Let the LLM decide source credibility | Explicit approved/excluded domain lists enforce editorial standards that LLMs cannot reliably self-police | ai-news-cafe-chatbot/README.md system prompt section |
KB type: Pinecone vector database (via AI Engine plugin) + embedded system prompt + local document collection Location in repo: knowledgebase/ (disambiguations, embeddings, guardrails dirs), AI News Cafe Knowledgebase/ (training data, prompt engineering, content sources, images) Estimated size: Hundreds of documents in AI News Cafe Knowledgebase/; Pinecone index with 3072-dimensional embeddings (text-embedding-3-small)
knowledgebase/
├── disambiguations/ # Journalism terminology
├── embeddings/ # Vector embeddings
└── guardrails/ # Editorial standards
AI News Cafe Knowledgebase/
├── Prompt Engineering/ # System prompts, build docs, copyright analysis
├── Training Data/ # Journalism training materials
├── Content Sources/ # Reference content
├── Images/ # Visual assets
├── Directories/ # Source directories
└── 2026 Employment Outlook/ # Industry outlook data
AI News Cafe Prompts/
└── .vscode/
└── ITI Claude Embedded ChatBot.md # Canonical system prompt
| Category | Files / format | Purpose | Update frequency |
|---|---|---|---|
| System prompt | PHP default + admin-editable text | Defines AI persona, journalism expertise, ethical guidelines, source lists | Per version update |
| Pinecone vector KB | 3072-dim embeddings via text-embedding-3-small |
RAG context for journalism-specific queries | Ongoing (AI Engine managed) |
| Journalism guardrails | GUARDRAILS.md, system prompt sections |
Accuracy requirements, source credibility, editorial standards | Per version |
| Training/reference data | Mixed formats (md, txt, csv, xlsx, pdf) in AI News Cafe Knowledgebase/ |
Background knowledge for embedding and prompt development | Development-time |
| Prompt engineering | Build docs, prompt iterations | Development reference for system prompt evolution | Development-time |
The knowledge base was assembled from curated journalism training materials, industry research (2026 Employment Outlook), credible source directories, and prompt engineering iterations. Documents were vectorized using OpenAI’s text-embedding-3-small model (3072 dimensions) and stored in a Pinecone index. The system prompt was iteratively refined through Cursor AI sessions, with each revision adding deeper journalism expertise, ethical frameworks, and source credibility controls.
System prompt approach: A comprehensive default system prompt embedded in the plugin (editable via admin UI) that defines the AI as a B2B media consulting expert with deep journalism knowledge, ethical AI standards, and curated source lists. Key behavioural guardrails: Accuracy requirements (must cite sources, flag uncertainty); source credibility filtering (100+ approved domains, explicit exclusion list for content farms); ethical AI journalism principles (transparency, accountability, human editorial control); privacy protections. Persona / tone configuration: Professional media consultant; authoritative but approachable; emphasizes evidence-based advice. Tool use / function calling: Pinecone vector search (RAG retrieval), Tavily web search (real-time information augmentation).
Developed through iterative Cursor AI sessions, evolving from a customized fork of the ITI gd-claude-chatbot architecture. The v2.0.0 release represented a significant stability overhaul — removing the dual AI Power integration, fixing database activation, adding comprehensive guardrails, and enhancing the admin UI.
| Phase | Approximate timeframe | What was built | Key commits or milestones |
|---|---|---|---|
| v1.x | Late 2025 | Initial chatbot with Claude + dual AI Power/AI Engine Pinecone integration | Foundation architecture based on gd-claude-chatbot |
| v2.0.0 | January 8, 2026 | Major stability release: removed AI Power, unified on AI Engine, database fix, admin settings tabs, guardrails system, UI/CSS enhancements | Breaking change release; 6+ focused update documents |
The CLAUDE.md file provides comprehensive project context for AI development tools. Development documentation includes detailed fix summaries (database, admin settings, UI, guardrails) that read as AI-session output. The cursor-chats/ directory contains exported development conversation histories. The plugin architecture follows ITI Shared Library patterns (singleton, RAG architecture, chat handler pipeline).
| Challenge | How resolved | Evidence |
|---|---|---|
| AI Power + AI Engine dual integration causing conflicts | Removed AI Power entirely in v2.0; unified on AI Engine-only Pinecone path | CHANGELOG-v2.0.0.md, PLUGIN-UPDATE-COMPLETE.md |
| Database tables not created on activation | Fixed dbDelta activation hook to properly create conversation log tables |
DATABASE-FIX-v2.0.0.md |
| Plugin activation crashing other plugins | Added PHP 8.0+ and WP 6.0+ version checks; defensive function_exists guards; singleton pattern | ai-news-cafe-chatbot.php (safety checks at top) |
| Stale v1 references in REST API endpoints | class-ainc-rest-api.php still references AINC_AI_Power_Integration and AINC_Dual_Integration in test/stats endpoints [CLAUDE NOTE: potential bug — classes were removed in v2.0] |
Code review of class-ainc-rest-api.php |
| Model / API | Provider | Role in product | Integration method |
|---|---|---|---|
Claude Sonnet 4 (claude-sonnet-4-20250514) |
Anthropic | Primary chat LLM — generates responses grounded in system prompt, RAG context, and web search | Direct API (https://api.anthropic.com/v1/messages) via class-ainc-claude-api.php |
| text-embedding-3-small | OpenAI | Query embedding for Pinecone vector search (3072 dimensions) | Direct API (https://api.openai.com/v1/embeddings) via class-ainc-pinecone-api.php |
| Pinecone Vector DB | Pinecone | Knowledge base storage and semantic retrieval | Direct API via class-ainc-pinecone-api.php; also available through AI Engine plugin |
| Tavily Search | Tavily | Real-time web search augmentation for current events and news | Direct API (https://api.tavily.com/search) via class-ainc-tavily-api.php |
| Tool | Category | Purpose |
|---|---|---|
| AI Engine (WordPress plugin) | RAG platform | Optional Pinecone integration and knowledge base management UI |
| WordPress REST API | Integration layer | Handles chat requests, connection testing, stats retrieval |
| Conversation log (MySQL) | Persistence | Stores multi-turn conversation history for context management |
| Tool | How used in build |
|---|---|
| Cursor | Primary development IDE; iterative code generation, debugging, documentation |
| Claude (via Cursor) | Architecture design, PHP class generation, system prompt development, guardrails definition |
| Version / Phase | Date | Summary of changes | Significance |
|---|---|---|---|
| v1.x | Late 2025 | Initial build: Claude chatbot with dual AI Power + AI Engine Pinecone integration | Foundation platform |
| v2.0.0 | January 8, 2026 | Breaking: removed AI Power; unified AI Engine; database activation fix; admin tabs; guardrails system; UI/CSS enhancements; safety on activation | Stability release; production-ready |
The v2.0.0 release represented a significant architectural simplification — removing the dual AI Power / AI Engine integration path in favor of AI Engine-only. This was driven by integration conflicts and maintenance burden. The guardrails system was added in v2.0.0 as a major new capability, reflecting growing emphasis on responsible AI journalism.
documentation/ directory is empty (docs live in plugin subdirectory and root-level markdown files)AI News Cafe Agents/ directory is empty (agent architecture referenced but not built)| Artifact | Path | Type | What it shows |
|---|---|---|---|
| Chatbot widget template | ai-news-cafe-chatbot/public/templates/chatbot-widget.php |
PHP/HTML | Chat interface layout |
| Chatbot CSS | ai-news-cafe-chatbot/assets/css/chatbot.css |
Stylesheet | Chat bubble styling, responsive design |
| Chatbot JS | ai-news-cafe-chatbot/assets/js/chatbot.js |
JavaScript | Chat interaction logic, streaming |
| Newspaper theme customization | WP Code/style.css, WP Code/header.php |
Theme files | Production site theme (Newspaper 12.7.3) |
| VS Code theme | AI News Cafe/.vscode/settings.json |
IDE config | Teal (#42e0be) brand color |
| Document | Path | Type | Status |
|---|---|---|---|
| Plugin README | ai-news-cafe-chatbot/README.md |
User guide | Complete (v2.0.0) |
| Quick Start | ai-news-cafe-chatbot/QUICK-START.md |
Getting started | Complete |
| Changelog | ai-news-cafe-chatbot/CHANGELOG-v2.0.0.md |
Version history | Complete |
| Guardrails | ai-news-cafe-chatbot/GUARDRAILS.md |
Ethics/accuracy spec | Complete |
| Guardrails Implementation | ai-news-cafe-chatbot/GUARDRAILS-IMPLEMENTATION.md |
Technical spec | Complete |
| Implementation Summary | ai-news-cafe-chatbot/IMPLEMENTATION-SUMMARY.md |
Architecture overview | Complete |
| Upgrade Guide | ai-news-cafe-chatbot/UPGRADE-GUIDE-v2.0.0.md |
Migration guide | Complete |
| Update Summary | ai-news-cafe-chatbot/UPDATE-SUMMARY-v2.0.0.md |
Release notes | Complete |
| v2.0.0 fix documents | Root-level *-v2.0.0.md files (6 docs) |
Focused fix summaries | Complete |
| Artifact | Path | Description |
|---|---|---|
| Plugin build spec | AI News Cafe Knowledgebase/AI News Cafe Plugin Build.md |
Original product requirements and architecture |
| System prompt reference | AI News Cafe Prompts/.vscode/ITI Claude Embedded ChatBot.md |
Canonical short system prompt |
| AI shortcode config | WP Code/AI Shortcodes.txt |
Production shortcode IDs for chatbot instances |
| Prompt engineering research | AI News Cafe Knowledgebase/Prompt Engineering/ |
Prompt iterations, copyright analysis |
| 2026 Employment Outlook | AI News Cafe Knowledgebase/2026 Employment Outlook/ |
Industry research data |
AI News Cafe emerged from Peter’s work at the intersection of B2B media and AI development. The observation that newsrooms were experimenting with generic AI tools — without domain-specific guardrails or knowledge bases — created an opportunity to build a purpose-built AI assistant for the media industry. The platform demonstrates that effective AI journalism tools require not just powerful models, but curated knowledge, ethical frameworks, and credibility controls. [CLAUDE NOTE: inferred from CLAUDE.md context and product positioning]
Research approach used: Industry analysis of B2B media AI adoption, review of journalism ethics frameworks, curation of credible vs. non-credible source lists. Key market observations:
What existing products got wrong: Generic AI chatbots treat all domains equally — they do not distinguish between credible journalism sources and content farms, do not enforce editorial standards, and cannot ground responses in a curated media industry knowledge base.
If B2B media professionals get an AI assistant that understands their industry, enforces journalism ethics, filters sources by credibility, and grounds responses in a curated knowledge base, they will adopt it as a daily decision-support tool — creating a defensible position in an emerging market.
The product evolved from a generic Claude chatbot fork (gd-claude-chatbot architecture) into a journalism-specific platform. v1.x established the core chat + RAG architecture with dual AI Power/AI Engine integration. v2.0.0 simplified the architecture (removing AI Power), added comprehensive guardrails, and elevated journalism ethics to a first-class product feature. The Newspaper theme integration and production deployment on ainewscafe.com demonstrate a full product-site pairing.
What works well: Claude-powered chat with journalism expertise; Pinecone RAG for knowledge-grounded responses; Tavily web search for real-time information; comprehensive guardrails system; responsive chat widget; multi-tab admin settings. Current limitations: REST API still references removed v1 classes (potential runtime errors on test/stats endpoints); documentation directory is empty (docs scattered across root and plugin dir); AI News Cafe Agents directory is empty (planned but not built); stale v1 references in some docs. Estimated completeness: v2.0.0 — production-deployed with live site; ~80% feature-complete relative to vision.
class-ainc-rest-api.php (test_connection, get_stats endpoints)AI News Cafe Agents/ directory with journalism-specific agent personasdocumentation/ directory_Manual input required — this section cannot be populated automatically._
| File / Path | What it contributed |
|---|---|
CLAUDE.md |
Sections 1, 2, 6, 10 — project overview, features, development context |
ai-news-cafe-chatbot/ai-news-cafe-chatbot.php |
Sections 1, 4, 7 — version, architecture, API integrations, safety checks |
ai-news-cafe-chatbot/README.md |
Sections 1, 2, 3, 4, 5, 7 — features, requirements, system prompt, guardrails |
ai-news-cafe-chatbot/CHANGELOG-v2.0.0.md |
Section 8 — version history, breaking changes |
ai-news-cafe-chatbot/GUARDRAILS.md |
Section 5 — ethical guardrails specification |
ai-news-cafe-chatbot/GUARDRAILS-IMPLEMENTATION.md |
Section 5 — guardrails technical implementation |
ai-news-cafe-chatbot/IMPLEMENTATION-SUMMARY.md |
Section 6 — architecture overview |
ai-news-cafe-chatbot/QUICK-START.md |
Section 9 — documentation artifact |
ai-news-cafe-chatbot/UPGRADE-GUIDE-v2.0.0.md |
Section 8 — migration documentation |
ai-news-cafe-chatbot/includes/class-ainc-claude-api.php |
Section 7 — Claude API integration details |
ai-news-cafe-chatbot/includes/class-ainc-pinecone-api.php |
Section 7 — Pinecone integration, embedding model |
ai-news-cafe-chatbot/includes/class-ainc-tavily-api.php |
Section 7 — Tavily web search integration |
ai-news-cafe-chatbot/includes/class-ainc-chat-handler.php |
Section 6 — chat pipeline architecture |
ai-news-cafe-chatbot/includes/class-ainc-rest-api.php |
Section 6 — REST API endpoints, stale references |
ai-news-cafe-chatbot/admin/class-ainc-admin-settings.php |
Section 4 — admin settings configuration |
ADMIN-SETTINGS-FIX-v2.0.0.md |
Section 6 — admin fix documentation |
DATABASE-FIX-v2.0.0.md |
Section 6 — database activation fix |
GUARDRAILS-ADDITION-SUMMARY.md |
Section 5 — guardrails addition context |
UI-CSS-ENHANCEMENTS-v2.0.0.md |
Section 6 — UI enhancement documentation |
INSTALLATION-INSTRUCTIONS-v2.0.0.md |
Section 4 — installation requirements |
PLUGIN-UPDATE-COMPLETE.md |
Section 8 — release completion status |
AI News Cafe Knowledgebase/AI News Cafe Plugin Build.md |
Sections 4, 10 — original build requirements |
AI News Cafe Prompts/.vscode/ITI Claude Embedded ChatBot.md |
Section 5 — canonical system prompt |
WP Code/AI Shortcodes.txt |
Section 9 — production shortcode configuration |
AI News Cafe/.vscode/settings.json |
Section 9 — brand color |
AI journalism tools have moved from experimental curiosity to operational necessity in under a year. FAYFO now claims to automate 70% of routine editorial tasks for newsrooms. Geneea’s RAG-based newsroom assistant serves major media houses. Timepath AI operates in 50+ newsrooms with 40,000+ storytellers. Perplexity Discover and its Comet browser deliver AI-curated news with citations to millions of consumers. The question is no longer “should newsrooms use AI?” but “which AI tools enforce the editorial standards that journalism requires?”
The vibe coding phenomenon amplifies both the opportunity and the threat. With AI-assisted development tools reaching $9.4 billion in funding and 82% developer adoption, building a journalism chatbot with RAG and source filtering is no longer a technical achievement — it is a weekend project. Anyone with a Claude API key, a Pinecone index, and a list of credible sources can replicate AI News Cafe’s basic architecture. The competitive moat cannot be the technology stack. It must be the editorial judgment embedded in the guardrails: the 100+ approved source domains, the explicit content farm exclusion list, the accuracy requirements, and the journalism ethics framework encoded in the system prompt.
LLM convergence creates a specific challenge for AI News Cafe. Perplexity, Feedly’s Ask AI, and Google Gemini all now offer search-grounded AI responses — the core value proposition of combining Claude with Tavily web search. But Perplexity’s academic audit found 72% of its citations were fabricated or inaccurate. Feedly synthesizes across articles but does not filter by editorial credibility. AI News Cafe’s differentiation lies in the claim that credibility-filtered, guardrail-enforced AI journalism assistance is categorically different from generic AI search. That claim becomes more defensible — and more necessary — as competitors demonstrate the risks of ungoverned AI in journalism contexts.
| Competitor | Category | Threat Level | Key Capability |
|---|---|---|---|
| FAYFO | B2B Newsroom AI SaaS | High | Fully automated newsroom workflows; replaces 70% of routine editorial tasks |
| Geneea Newsroom Assistant | B2B Media AI | High | RAG-based newsroom assistant for research, tagging, related content |
| Timepath AI | Newsroom-native AI | Medium | Task-specific editorial modules; 50+ newsrooms, 40,000+ storytellers |
| Perplexity Comet | AI Browser + News | Medium | Embedded AI assistant in Chromium; news briefings with cited sources |
| VerifyAI (WordPress) | Fact-checking plugin | Medium | One-click WP editor fact-checking; dual AI support (GPT-4o / Gemini) |
| GeoBarta | AI News Summarization | Low | Clustering technology, ad-free consumer summarization |
| Feature | Who Shipped It | AI News Cafe Status |
|---|---|---|
| AI-curated news feed with topic tabs + TTS audio | Perplexity Discover | Not built — raises bar for news delivery format |
| Ask AI article synthesis (25 articles, inline citations) | Feedly | Not built — direct competitor to our RAG chat |
| Bulk article selection + AI prompts | Feedly | Not built — multi-article analysis gap |
| Conversational news personalization | NewsBreak (NBot) | Not built — consumer conversational assistant |
| AI notification summaries with source transparency labels | Apple News (iOS 26) | Not built — sets user expectations for disclosure |
| One-click content fact-checking in WP editor | VerifyAI plugin | Not built — adjacent WordPress capability |
| Automated newsroom workflows (70% task replacement) | FAYFO | Not built — fundamentally different scope |
| Differentiator | Status | Detail |
|---|---|---|
| RAG-grounded journalism responses | Partially eroded | Geneea uses RAG for newsroom assistant; Feedly Ask AI synthesizes from curated sources |
| Real-time web search integration | Eroded — now table stakes | Perplexity, Feedly, and Geneea all integrate real-time search |
| WordPress-native AI journalism chatbot | Still unique | No WP chatbot competitor with journalism guardrails exists |
| Source credibility filtering at prompt level | Still unique | Ground News has consumer bias ratings; no one enforces credibility at the LLM prompt level |
The roadmap prioritizes cleaning up v2.0 technical debt first, then ships the features that leverage AI News Cafe’s unique guardrails infrastructure — an asset no competitor has.
Tier 1 — Critical (Next Build Cycle)
Tier 2 — High Value (Near-Term)
Tier 3 — Strategic (Medium-Term)
Prioritization rationale: Technical debt cleanup is Tier 1 because the stale v1 class references in the REST API are a potential runtime error in production. Transparency labels are Tier 1 because Apple News has set user expectations and Perplexity’s citation fabrication scandal makes disclosure a market differentiator. The journalism personas fill the empty AI News Cafe Agents/ directory that was planned but never built. The claim verification pipeline (Tier 2) is the product’s most defensible feature opportunity — it combines all three APIs (Claude + Tavily + Pinecone) in a verification-first pipeline that directly addresses Perplexity’s 72% fabrication problem.
| Skill | What It Enables |
|---|---|
news-credibility-scoring |
Score news sources on credibility using ownership analysis, editorial standards assessment, correction history, and domain authority. Directly supports the Tier 1 credibility visualization and Tier 2 bias analysis features. |
multi-agent-journalism-workflow |
Orchestrate multi-agent journalism workflows where specialized agents collaborate with audit trails. Supports the Tier 1 journalism personas and Tier 3 multi-agent orchestration. |
ai-journalism |
Ethical AI-assisted journalism practices, news curation, and credible source integration. Provides the domain framework for all journalism-specific features. |
fact-checking |
Professional fact-checking methodology for verifying claims, assessing source credibility, and ensuring content accuracy. Underpins the Tier 2 claim verification pipeline. |
Strengths:
Gaps we’re honest about:
AI News Cafe Agents/ directory has been empty since v2.0 launch — the agent architecture was planned but never builtWhat we’re watching:
Portfolio context: AI News Cafe demonstrates ITI’s ability to apply AI to journalism — a domain where credibility, ethics, and source quality are not features but prerequisites. The product’s value as consulting portfolio evidence lies in showing that responsible AI journalism requires guardrails architecture, not just model capability. In a market where Perplexity fabricates 72% of citations, the case for credibility-enforced AI is becoming stronger, not weaker.