An AI agent that hallucinates, misses security threats, or generates unexplainable outputs isn’t an asset—it’s an operational liability.
As enterprise AI matures beyond flashy chat interfaces, the engineering challenge has shifted dramatically: How do we build autonomous AI agents that are reliable, audited, and production-ready?
That exact question sits at the center of the “Forge the Future” Hackathon hosted by Elastic and backed by AWS, in partnership with Nasscom AI, Sarvam AI, and HackCulture. Following weeks of proposal submissions and team formations, shortlisted developer teams gathered in Bengaluru for the grand finale—competing not just for a ₹600,000 prize pool and Sarvam AI credits, but for a spotlight at Elastic{ON} Mumbai.
Here is my engineering breakdown of what this hackathon reveals about the state of enterprise AI in India, and why production readiness is the new benchmark for builders.
The Four Tracks: Testing Real-World Enterprise Utility
Instead of generic “build any AI tool” prompts, the hackathon forced developers to execute solutions directly on the Elasticsearch Platform hosted on AWS cloud infrastructure:
| Track Focus | What Teams Built | Real-World Impact Target |
| 1. Intelligent Search & AI Platforms | Hybrid search engines outperforming standard vector DBs, RAG agents for BFSI, and multi-step ticket triage flows. | Replacing raw vector lookups with context-aware, domain-specific retrieval. |
| 2. Observability & Autonomous Ops | Agents leveraging ES|QL (Elasticsearch Query Language) to analyze log streams, correlate traces, and suggest fixes. | Reducing Mean Time to Resolution (MTTR) with self-healing system workflows. |
| 3. Security & AI-Driven SOC | Automated CERT-In compliance checkers, BFSI fraud detection engines, and ES|QL-powered threat hunting. | Transforming reactive SOC dashboards into proactive, agentic defense hubs. |
| 4. Vertical Industry Solutions | Explainable AI models for BFSI risk assessment, e-commerce/retail intelligence, and health insurance auditing. | Providing clear audit trails for highly regulated industries. |
Blogger’s Insight: Notice how every single track prioritizes explainability and auditability. In financial services, healthcare, or cybersecurity, an AI model that cannot explain why it flagged a transaction or system anomaly is instantly rejected by compliance teams.
From Bengaluru Code to the Elastic{ON} Stage
The journey began back in July with open registrations. Following code submissions and evaluation rounds, selected teams advanced to the in-person Bengaluru finals, working side-by-side with solution architects from Elastic and AWS.
The Incentive Structure
- 1st Place: ₹300,000 Cash + ₹75,000 Sarvam AI credits
- 2nd Place: ₹200,000 Cash + ₹50,000 Sarvam AI credits
- 3rd Place: ₹150,000 Cash + ₹25,000 Sarvam AI credits
Beyond the financial rewards, the top three teams won a far bigger prize: a direct ticket to Elastic{ON} Mumbai at the JW Marriott Sahar. There, winners presented their production-ready agents directly to enterprise CTOs, SRE leads, and security directors.
Why “Agentic AI” Needs Enterprise Infrastructure
For the past two years, the AI ecosystem was dominated by standalone LLM wrappers that looked impressive in 30-second Twitter demos but broke under real production loads.
What makes the Elastic + AWS stack so strategic for this challenge is the underlying architecture:
- Hybrid Search Over Pure Vectors: Pure vector databases struggle with exact keyword matching, metadata filtering, and real-time indexing. Combining lexical search with vector embeddings gives agents accurate context retrieval.
- ES|QL Query Performance: Using pipe-structured query languages enables agents to run complex data transformations across massive log volumes without custom pipeline code.
- Multilingual Context via Sarvam AI: Integrating Indian-language foundational models allows developers to build local enterprise applications tailored to regional user bases.
Final Thoughts: The End of “Demo-Only” AI
The Forge the Future hackathon represents a welcome shift in the developer landscape. The Indian AI scene is moving away from toy projects and wrapping APIs into UI templates. Instead, developers are tackling hard, back-end infrastructure problems: log correlation, regulatory compliance, and auditable search.
If your AI solution can’t survive real-world data noise, enterprise latency limits, and strict security compliance, it isn’t ready for production.
What’s Your Take?
Are you building with agentic AI models in your current tech stack? What is your biggest hurdle when transitioning agents from local prototypes to production—latency, cost, or explainability? Drop your thoughts in the comments below!