Speak at The Fifth Elephant 2026 Annual Conference

While working with our customers in the MSME sector, we realized that most of them rely on WhatsApp for one important reason. It is really easy to capture and send images - invoices, receipts, product photos, attendance, screenshots, etc. A lot of useful business information is trapped inside these images. The problem is that each image is unique and could contain different kinds of information. It is difficult to extract and use this data meaningfully.

This talk explains how we built a data ingestion and processing pipeline that handles different kinds of media while performing OCR and data extraction. We use real customer use cases to explain how messy inputs are converted to useful business data. We will talk about how we made the pipeline reliable and increased customer confidence by building readiness checks, audit trails, replays, and safe downstream processing.

This talk presents an interesting LLM tool pattern to extract structured data from unstructured inputs when the schema is not known in advance. This is useful when the shape of the information changes from one input to the next. This pattern is what enables us to use the same pipeline for all incoming multimedia messages.

We will also look at what happens after extraction - how the extracted data is used in the downstream search and automation workflows, where semantic and keyword search help, how techniques like Reciprocal Rank Fusion (RRF) and reranking improve search quality, and how a simple observability layer helps us with multi-tenant metering and understand cost, accuracy, and answer quality in production.

Bio

Satyajeet Jadhav is the co-founder of Querygen.ai, where he works on AI automation for MSMEs. He has 15 years of experience across software engineering, customer success, and building SOCs. Before Querygen.ai, he worked at startups including Last9, Simplilearn, and CleverTap. He also built thinkdeli.com. - https://www.linkedin.com/in/satyajeetjadhav/