The Fifth Elephant 2026 Annual Conference

31 July 2026 | NIMHANS Convention Centre, Bengaluru

Built for humans. Now rebuilding for agents.

The Fifth Elephant 2026 brings together practitioners who have already hit the walls of building for agents in production — and rebuilt from first principles, with the numbers to prove it.

Get your ticket → https://hasgeek.com/fifthelephant/2026/#tickets
Full schedule → https://hasgeek.com/fifthelephant/2026/schedule

Who is attending the conference on 31 July

Founders, principal architects, engineering managers, division heads, and data scientists — from Razorpay, Microsoft, Cleartax, Thomson Reuters, Target Corporation, Sequoia, Zoho, e6data, Opella, and IIIT Bangalore, among others.

Keynote

Rethinking data architecture for the agent era

Sunny Bains (PingCAP)

The comfortable assumption is that an “AI-ready” database just needs a vector index bolted on. In practice, that leads to fragmented state across object storage, relational metadata, and vector databases — none designed to talk to each other at agent scale.

Next-generation databases have to be agent-first, not human-first with an AI feature added on.

Track highlights

Track 1: Data Engineering & Infrastructure

90+ submissions, distilled to 24 talks. This track is about what it actually takes to run data platforms at scale, and the trade-offs associated with technology choices.

Highlights:

• Amit Prabhu (Razorpay) — cut denormalization compute cost and generation time by 85%+ building a Lakehouse at India scale
• Sujeet Gholap (Oodle AI) — building an observability platform on S3 + Lambda, hard engineering on hard problems
• Mayur Jadhav (OrcaSheets) — giving business users lake-scale outcomes from their laptop, without the data engineering overhead

Curated by Piyush Goel (Razorpay), Ravi Balgi (Datanimbus), and Ranganadh Thata (Mico)

Track 2: Agentic AI — Platforms and Use Cases

Agents are easy to demo. Hard to run. This track is about what happens after the prototype — real users, real latency, real failure modes the demo never warned you about.

Highlights:

• Dhruv Nigam (Dream 11) on why voice agents built as “chatbot + STT + TTS” break in production, and the native voice architecture that got Googly Bhai to 5M+ daily users at ~2-second latency
• Vivek Kalyanarangan (IDfy) on cutting ML inference cost 10× across a 40+ model fleet via quantization
• Dr. Karthika Vijayan (Sahaj Software) on why multi-agent consistency failures are usually a language-understanding problem, not an orchestration one

With more talks from Atlassian, Microsoft, IBM Research, and PhonePe.

Curated by Yash Gandhi (OrcaSheets), Jagadish Kasi (Tryft), and Ramakrishna Reddy Y (Red Hat)

Birds of a Feather (BOF) sessions

BOFs are where the real production war-stories come out — informal, practitioner-led. Six sessions on the schedule:

1. Agentic AI: Path to Production — architecture, evaluation, and operational trade-offs for agents beyond the prototype stage. Led by Karthika Vijayan (Sahaj Software)
2. Things We Wish We Knew Before Building Agentic Systems — the assumptions that turned out wrong, and what we’d redesign today. Led by Sujit Kamthe (Sahaj Software)
3. Beyond SWE-bench: How Do We Evaluate AI Agents for Real-World Workflows? — building evaluation criteria and datasets when no benchmark exists, and connecting offline evals to online outcomes. Led by Shashank Rao (Atlassian)
4. Engineering Agent Quality Loops (White-Box) — turning traces, logs, and evals into real engineering decisions when an agent’s path — not just its answer — goes wrong. Panel: Prateek Mandloi (Atlassian), Sunny R Gupta (Metaforms)
5. Designing Reliable Agentic Loops — component-wise evaluation of agent loops, and turning failed runs into regression tests. Led by Priyanga (nilenso), with Akash Sathish (Sahaj)
6. Build vs. Buy — Led by Ramakrishna Reddy Y. Details upcoming.

Drop into any BOF that matches what you’re wrestling with — no sign-up needed.

Demo table - two voice bots, two very different production problems

Googly Bhai (Dream11) - Dhruv Nigam brings the real-time voice architecture behind the AI sports-streaming companion that hit 1.1M concurrent viewers this IPL season. Talk to it, push its limits, and ask Dhruv what breaks. (Details on suggested prompts to arrive EOD.)

Sarlaben (AmulAI, The Flywheel) - Sarlaben is Amul’s production voice and chat helpline for Gujarati dairy farmers, answering questions on animal health, breeding, and government schemes. At the table, talk to it live over a demo web UI in Gujarati or English — a typical exchange takes a minute or two. Gautam Rajeev and Chakshu Gautam will be there to walk you through the architecture.

Things to try:

• Ask about a common cattle ailment or symptom
• Ask about breeding cycles or timing
• Ask about a government dairy scheme
• Switch mid-conversation between Gujarati and English
• Ask something it doesn’t know — see how it handles not knowing

Installation: Museum of Human Practice

Expo area, NIMHANS Convention Centre

As AI takes over more of our work, we don’t just change how tasks get done — we change what we repeatedly practise. This self-guided installation invites you to pin activities (writing SQL, giving hard feedback, planning a trip) onto one of three boards: things we’ve gladly delegated, things we wish we could delegate, and things we hope to preserve.

Add your own card. There’s no correct answer — the same activity often ends up on all three boards. By end of day, it becomes a collective snapshot of how a room full of AI practitioners actually thinks about delegation and human capability.

Pick up a badge card on your way in: “Today I’m practising ___.” Fill it in, clip it on, and let it start conversations beyond the booth.

By Priyanka Bharadwaj, who works on human connection in the age of AI. More about the installation →

Meet the authors

Dr. Prasanna Bhogale — An Unfriendly Introduction to Causal Inference
Community member, speaker at two past Fifth Elephant editions. The shortest serious route from a causal question to a number, with no verbose hand-holding.
📖 ~30 copies on-site · ₹999 · Book details

Vivek Kalyanarangan — Quantization and Fast Inference (Manning)
Speaking on Track 2 (Beyond GPUs — cutting ML inference cost 10×). His book covers the full quantization pipeline, from first principles to production deployment on CPUs, edge, and mobile.
📖 30 copies on-site, with a thank-you note in every one · Book details · 45% off all Manning books for conference attendees — grab the code at the venue

Both authors will be at the venue — come say hello and take a copy home.

Get your ticket to attend

Your ticket covers the Fifth Elephant 2026 season:

• Main conference day — 31 July, NIMHANS Convention Centre
• Post-conference workshop — 8 August, Sahaj Software (two sessions: Memory in Agentic Systems, The AI Engineer’s Playbook)
• The Fifth Elephant Winter edition — November 2026 (Bangalore + virtual)

🎟 Conference tickets and annual memberships

Cancel any time up to 24 hours before the event for a full refund — this doesn’t apply once you’ve attended a workshop under the ticket.

Contact & sponsorship

Sponsorship inquiries: sales@hasgeek.com
📞 Call or WhatsApp: +91 76763 32020
📧 info@hasgeek.com