This article is a redacted version of the presentation delivered by Jiri Hana from our partner Geneea at the Newsroom Summit 2025, organized by WAN-IFRA in Copenhagen on November 18–19, 2025. This summary aims to capture the key insights shared during the event.
Anyone who has ever conducted an interview knows the real work begins when the recorder stops. The conversation might have been brilliant, but now comes the time-consuming reality: hours of audio to review, transcribe, structure, and polish into a coherent article.
What if AI could handle the heavy lifting while you focus on what matters: the story itself?
The JEDI Project is an initiative designed to transform raw interview recordings into near-publishable articles with minimal human oversight. Unlike traditional transcription services that convert speech to text, JEDI produces structured, coherent narratives that maintain journalistic quality and editorial standards.
Why We Built JEDI
Newsrooms today face a familiar challenge: shrinking budgets, tight deadlines, and the constant pressure to produce quality content. Transcription and article drafting can consume 3-4 hours for every hour of recorded interview. That’s time journalists could spend reporting, investigating, or conducting more interviews.
JEDI aims to reclaim that time. It processes interview recordings of up to 60 minutes and generates articles between 2,000 and 3,000 words, complete with proper structure, tone, and formatting that match editorial expectations. The system is trained on English-language content and has been rigorously tested to ensure quality output.
How It Works: Three Intelligent Phases
JEDI’s workflow guides your content through three progressive stages, each designed to refine and enhance the material:
Phase 1: Transcript Cleanup
Raw transcripts are messy. They’re filled with fragmented dialogue, unclear phrasing, and grammatical inconsistencies. JEDI’s first phase focuses on hygiene: merging dialogue segments, handling split sentences, correcting grammar and typos. Every change is verified against the original transcript to ensure accuracy and prevent hallucinations.
Phase 2: Article Creation
Once the transcript is clean, JEDI drafts the article. It structures the content into topic-based sections, settles the interview into a narrative flow, and generates follow-up questions to fill gaps or add depth. If the article exceeds the target word count, it intelligently breaks content into smaller, focused parts. The system can also address dialogue that’s condensed or unclear, expanding it for better reader comprehension. All while staying within the 3,000-word limit.
Phase 3: Refinement and Verification
The final stage is where JEDI ensures publication readiness. It checks terminology consistency, adjusts formatting, reviews sentence structure, and extracts factual claims to verify them against the original transcript. This built-in fact-checking mechanism provides traceability for both editors and journalists, linking every statement back to the source audio.
Built for Real Newsroom Workflows
JEDI is powered by a robust technical stack using Python and advanced Large Language Models, including OpenAI, Mistral, and Claude. The final output is automatically converted into Google Docs format, making it easy for editors to review, comment, and collaborate.
The system has been tested extensively on English-language interviews, with a strong focus on maintaining journalistic integrity and editorial quality. While JEDI dramatically reduces manual labor, it’s designed to work with journalists, not replace them. The human eye remains essential for final review and editorial judgment.
What Makes JEDI Different
Traditional AI transcription tools stop at text conversion. JEDI goes further by understanding context, structure, and narrative flow. It doesn’t just transcribe, it authors. The system produces drafts that read like a journalist wrote them, not a machine.
More importantly, JEDI is transparent. Every claim in the generated article can be traced back to the original recording, ensuring accountability and making fact-checking straightforward. This is critical for newsrooms where accuracy isn’t optional. It’s foundational.
Join Our Beta Program: Completely Free
We’re opening JEDI to a select group of newsrooms for beta testing. This is your opportunity to experience firsthand how AI can transform your editorial workflow, reduce production time, and free your team to focus on what they do best: journalism.
The beta is completely free. We’re looking for newsrooms willing to test JEDI with real interviews, provide feedback, and help us refine the system for the specific needs of editorial teams.
What you’ll get:
- Full access to JEDI for your newsroom during the beta period
- Direct support from our technical team
- The chance to shape a tool built specifically for journalists
- Significant time savings on transcription and article drafting
What we’re looking for:
- Newsrooms conducting regular interviews
- Teams willing to provide honest feedback on output quality and workflow integration
- Editorial environments where AI-assisted content creation can be properly evaluated
- Access to raw interview audio files that will help us optimize the model for diverse journalistic contexts
Ready to Transform Your Workflow?
If your newsroom is interested in participating in the JEDI beta program, we’d love to hear from you. Sign up through our beta registration form, and our team will reach out to discuss implementation and onboarding.
The future of interview-to-article production is here. Fill in this form and join us in testing it!
About This Project
The JEDI Project is a collaboration between Atex and Geneea, made possible thanks to the FAIR EU Fund. Atex has been selected among the winners of FAIR’s Spoke 5 grant, focused on High-Quality AI. The FAIR initiative is promoted by the Italian Ministry of University and Research (MUR) and funded by the European Union – NextGenerationEU.
Geneea has been building AI solutions for the news-media industry since 2014. Based in Prague, the company specializes in natural-language processing and generative AI that help publishers automate routine tasks and gain deeper insights into their content. Its technology spans text and audio analysis, metadata enrichment, and automatic article generation, and is used by major media organizations such as Mediahuis, The Atlantic, RTS, The Hindu, and the Czech News Agency. Geneea’s mission is to support newsrooms with practical, reliable AI tools that streamline workflows and let journalists focus on storytelling.
This funding and partnership enable us to develop JEDI as a tool specifically designed for journalism, ensuring that AI technology serves the media industry’s unique needs while maintaining the highest standards of quality and accuracy.