When it comes to AI in newsrooms, I’ve seen two extremes: tools that promise to change everything but end up being little more than a flashy novelty, and tools that quietly make a real difference.

At Newslaundry, we knew which kind we wanted to build.

We didn’t need another AI anchor or automated content generator. What we needed was something far simpler: a tool that solved a problem for our teams, freed up time, and helped us work smarter.

That’s how Ask Birubala came to life.

Understanding the problem

Like many small, independent newsrooms, we run lean. Every team member wears multiple hats. 

Our product and subscription teams often spend hours answering repetitive subscriber queries, explaining workflows to new team members, or troubleshooting technical issues. Senior developers and product managers are asked support questions far more often than they should be.

Every hour spent on repetitive tasks is an hour taken away from improving our product, analysing our subscription performance, or experimenting with new growth strategies.

What we needed was a single source of truth; a tool that could answer recurring product and technical questions, help our support team respond to users quickly, and onboard new people without always relying on senior team members.

Enter AI 

Ask Birubala is an AI-powered retrieval-augmented generation (RAG) tool that does two things well:

• Answers product and technical questions: New team members (or even external partners) can ask about our subscription management software, payment gateway integrations, or publishing workflows — and get clear, contextual answers.

• Drafts responses to subscriber queries: When a user writes in with an issue, our support team can paste the e-mail into Ask Birubala. The tool generates a ready-to-send, step-by-step response explaining what happened and how to fix it.

We’ve deliberately kept a human in the loop. The tool drafts, but a team member reviews and personalises the response before it’s sent.

In short, Ask Birubala acts like that super-knowledgeable colleague who has all the answers — and also drafts your e-mails for you.

How it has helped

The results have been immediate and tangible:

• We’ve saved 10+ hours per week for senior developers and product managers.
  

• Team members who earlier spent all their time on support queries are now upskilling into data analysis and campaign management.
  

• Job satisfaction has gone up because repetitive, low-impact work has gone down.
  

One colleague summed it up perfectly: “It feels like we finally have a living, breathing manual that actually answers our questions.”

The tech behind it

Ask Birubala runs on a modular, easy-to-iterate stack:

• Backend: NodeJS app using Langchain, OpenAI, and Pinecone for vector embeddings.
  

• Backend architecture: Data is stored in PDFs, split into chunks, vectorised, and stored with metadata (like topic and platform tags) for context-specific responses.
  

• Frontend: A lightweight React app on Vercel with a simple, intuitive chat UI.
  

• Documentation: We use Scribehow to convert screen recordings into PDFs for the knowledge base.
  

This setup makes it easy to continuously update the knowledge base as our workflows evolve.

The next big step for Ask Birubala is to use AI not just for support and workflows, but also for insights. We’re exploring ways to analyse subscription data through the tool and surface actionable insights for our editorial team — helping them make more informed decisions about content, coverage, and reader engagement.

Imagine a reporter being able to ask, “What kind of stories have the highest conversion rates among new subscribers?” and getting an instant, data-backed answer. That’s where we’re headed.

The big lesson

What if we could build one tool that answers all our recurring product, tech, and subscriber questions so our senior developers and PMs can focus on more meaningful work?

Throughout the CUNY Journalism AI Lab Fellowship, this idea became a proof of concept. It wasn’t flashy, but it addressed a real pain point. And that was the goal.

With the incredible engineering work of Rishabh Dixit, we turned that MVP into a tool embedded in our daily workflows — and we’re just getting started.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .