Speak at The Fifth Elephant 2026 Annual Conference

We built a production multi-agent assistant for a complex enterprise planning workflow where AI assists, rather than replaces, human decision-making. Users express their intent through natural language, the system extracts structured semantic information for human review, and multiple specialist agents collaborate to retrieve data, perform analysis, and generate recommendations. Our initial architecture followed a familiar orchestrator-driven design. The first production incidents taught us something unexpected: every individual agent behaved correctly, yet the overall system did not. We assumed orchestration would keep multiple agents consistent. Production taught us that consistency is established before orchestration begins. Ambiguous user requests were interpreted differently across agents, conversational context caused semantic state to drift, and independently correct outputs became collectively inconsistent. Retrieval operated on different assumptions, downstream reasoning diverged, and response generation was left reconciling conflicting interpretations. We realized we were treating consistency as an orchestration problem when it was fundamentally a language understanding problem.

The solution was an architectural pivot inspired by lessons from traditional NLP rather than a more sophisticated orchestration. We moved language understanding from an implementation detail inside the orchestrator to a shared architectural capability that every agent depends on. Intent understanding, entity grounding and disambiguation and state-transition-aware conversation history/context management establish a canonical semantic state before orchestration begins. The extracted entities became both the contract between downstream agents and the UI through which human planners interact with the system. Rather than presenting another orchestration framework, this talk shares the production failures that drove this redesign, the engineering trade-offs behind it, and why decades-old NLP concepts including intent, entities, and conversational state - remain one of the strongest foundations for building reliable production multi-agent systems in the era of agentic AI.

Key takeaways

• Learn why consistency failures in production multi-agent systems are often natural language understanding (NLU) problems, and see how classical NLP concepts can be repurposed as architectural building blocks for modern agentic AI.
• Understand how to establish a shared semantic state that keeps downstream agents consistent, even across ambiguous and multi-turn conversations.

Target audience

• AI/ML engineers building production-grade LLM and multi-agent applications, aiming at deploying reliable, scalable AI systems in production.
• Solution architects designing enterprise AI systems and agentic workflows.
• NLP practitioners interested in applying classical language understanding techniques to modern agentic architectures. Practitioners looking for pragmatic, engineering-first solutions to production challenges.
• Developers building conversational assistants and human-in-the-loop decision support systems.

Speaker bio
Dr. Karthika Vijayan is a Solution Consultant at Sahaj Software. She has been conducting research in the field of conversational AI with voice and text data for almost a decade. Her research has been published in several journals and presented at various international conferences. Prior to joining Sahaj Software, she worked as a research fellow at the National University of Singapore and at IISc Bangalore. She has done her PhD from IIT Hyderabad.

Previous talk links

1. Fifth El Bangalore (2025): https://www.youtube.com/live/4Nw89rICm-s (6:09:55 - 6:47:30)
2. Devday Pune (2025): https://youtu.be/1JSZE-1Zlvg?feature=shared
3. WiDS (2025): https://www.youtube.com/watch?v=_es29z32_Xs (1:26:40 to 2:02:10)
4. Devday Bangalore (2025): https://youtu.be/WQUt5Sjf5U8?feature=shared

Profile links:
https://scholar.google.com/citations?user=fJp6O0UAAAAJ&hl=en
https://www.linkedin.com/in/karthika-vijayan/
https://www.researchgate.net/profile/Karthika-Vijayan