

Ask Birubala is an AI-powered RAG (Retrieval-Augmented Generation) tool by Newslaundry – an in-house expert, always ready with the right answer, built to simplify subscription support workflows and act as a reliable assistant for all product and technical queries across the organization.
- A troubleshooting assistant: For those head-scratching subscriber issues, Birubala steps in to quickly identify problems and guide resolutions.
- A knowledge assistant: For onboarding new team members or collaborating with external partners. Birubala acts as a comprehensive knowledge base, reducing the learning curve.
- A documentation shortcut: No more endless searching for answers! This tool drastically cuts down the time spent digging through various documents.

“Ask Birubala takes its name from Newslaundry’s brand mascot, Birubala — seen here with her partner-in-crime, Birubal.”
We spoke to Chitranshu Tewari, Director, Product and Revenue, about this project that was conceived and developed as a proof of concept during Chitranshu’s participation in the 2024 CUNY AI Journalism Lab Fellowship, a global program hosted by the Craig Newmark Graduate School of Journalism.
The tool was then developed by him and Newslaundry’s senior developer, Rishabh Dixit.
The two core pillars of Birubala
Ask Birubala excels in two critical areas, making it an indispensable asset for Newslaundry’s teams:
1. Answering technical and product queries
Ever wondered how to map bulk subscriptions or troubleshoot a failed payment? Birubala has the answers. It’s a go-to resource for anything related to Newslaundry’s internal product stack, subscription workflows, or publishing systems. This feature is particularly invaluable for new hires and support teams who need to quickly get up to speed on internal processes. Imagine the time saved when a new joiner can instantly learn:
- How to map bulk subscriptions
- How to troubleshoot failed payments
- How to create and track campaigns in our CRM
2. Troubleshooting subscriber issues and drafting email replies
This is where Birubala truly shines in the realm of customer support. By simply pasting a user’s support email into the tool, Birubala can generate a clear, context-rich response. This isn’t just a generic reply; it explains what went wrong, what action the user needs to take, and why the issue occurred.
It achieves this level of detail through meticulous training on years of historical support emails, internal FAQs, and insightful notes from product managers and developers.
A single source of truth
The overarching goal behind Ask Birubala is ambitious yet vital: to create a single source of truth for all technical, operational, and support knowledge within Newslaundry. This centralized approach ensures consistency and accuracy across all interactions.
Who benefits?
Teams across the entire organization can leverage Birubala’s capabilities to:
- Retrieve technical answers quickly: No more waiting for a senior developer to be available.
- Write accurate and consistent responses to support queries: Ensuring a unified voice and correct information.
- Onboard new team members with minimal handholding: Accelerating productivity from day one.
- Keep internal documentation centralized and useful: Making sure knowledge is always accessible and up-to-date.
Features and tech
Ask Birubala isn’t just smart; it’s smartly built. Here’s a look at its key features and the technology powering it:
- Source referencing: Transparency is key. Birubala displays citations, showing exactly where the information came from.
- Live knowledge base editing: The system is designed to continuously improve. Users can revise or correct documents directly within the tool.
- Topic and scope filters: Need to narrow down your search? Select from categories like “Tech Stack” or “Subscriber Issues” to focus your results.
- Contextual conversations: Birubala remembers! It maintains thread context to improve the accuracy and relevance of follow-up responses.
Technical stack:
- Framework: Next.js
- Frontend: A React app hosted on Vercel, built with shadcn/ui for a clean and consistent interface.
- Backend stack: A Node.js app utilizing Langchain, OpenAI, and Pinecone for vector operations. This setup supports long-term memory and citation in responses.
- Backend architecture: Documentation is stored as PDFs (generated from screen recordings via Scribehow). These PDFs are split into chunks, vectorized, and stored in Pinecone, each tagged with metadata like “Topic” (e.g., subscription, paywall) and “Platform” (e.g., payment gateways, CRM tools). When a query is entered, it’s converted into an embedding, relevant document chunks are retrieved, and a contextual, cited response is generated.
- AI integration: OpenAI serves as the language model provider, with Vercel AI SDK managing interactions.
- Vector database: Pinecone is the chosen vector database.
- Documentation: Created using ScribeHow, which efficiently converts screen recordings into digestible PDFs for ingestion.
Overcoming the hurdles
The journey to building Ask Birubala wasn’t without its challenges, highlighting the dedication of the Newslaundry team:
- Building the knowledge base: A significant hurdle was the lack of detailed documentation. Many processes were undocumented or manual, making the compilation of reliable internal documentation a slow but crucial first step.
- Learning AI concepts from scratch: The team embarked on a steep learning curve, understanding the fundamentals of AI and LLMs while simultaneously building the tool.
- Keeping the tool updated: Newslaundry’s tech stack and workflows are constantly evolving, requiring ongoing updates to the knowledge base and regular re-indexing in the vector database.
Iterations and improvements
Newslaundry’s commitment to Ask Birubala is ongoing. Continuous improvements include:
- Reviewing AI response quality: Regularly assessing and refining the accuracy of the AI’s answers.
- Introducing a feedback mechanism: Users can rate answers and suggest corrections, ensuring the tool gets smarter with every interaction.
- Classifying content in pinecone: Using metadata for better content organization has made retrieval faster and more accurate.
- Enabling knowledge base edits from within the interface: This crucial feature allows for on-the-go improvements to the tool’s knowledge.
The impact: measurable success
The implementation of Ask Birubala has already yielded significant and measurable positive impacts:
- Time savings: Senior developers and product managers are now saving more than 10 hours per week — time previously spent manually fielding technical queries.
- Reduced support load: Fewer team members are needed for basic support queries, allowing those who previously handled support to transition into more strategic roles like data analysis and campaign management. This shift has led to:
- Improved employee productivity
- Better return on investment for each role
- Higher job satisfaction, especially in positions that historically saw high attrition.
“Ask Birubala has saved us over 10 hours a week and freed our team to move from repetitive support work to data analysis and campaign management.”
Ask Birubala isn’t just an internal tool; it’s a testament to Newslaundry’s innovative spirit and commitment to operational excellence. By leveraging AI, they’ve not only streamlined internal processes but also fostered a more productive and satisfying work environment for their teams.