Panel Discussion: AI Needs a Foundation That Never Fails
summary
Enterprises report production agents that classify and extract pages from 20–30‑page applications, run signature‑verification models on lakhs of documents daily, and reduce TAT from 30–40 days to 2–3 hours. They deploy Gemini 3 GSUs running locally in India, add fallbacks via orchestration and open‑weight models, and label agent autonomy as A0–A3. Hard limits are data silos, Kafka replication latency, token economics for LLMs, the need for vector stores/feature stores, and auditability and explainability for regulated decisions.
AI-written summary · about
chapters (11)
- 0:00AI foundation panel intro
- 2:14Audience poll on AI impact
- 4:40Panelist AI use cases
- 9:12Business gains from AI
- 11:18Document processing in AI
- 17:39Scaling AI to production
- 22:33Data silos in AI adoption
- 25:41Data challenges in AI-native firms
- 29:56AI-native enterprise definition
- 33:07AI system architecture evolution
- 38:01Closing remarks on AI platforms
related talks
24mScalable GenAI in Production
38mUnlocking the Future: GenAI, DevOps, and Fintech Innovation
21mOvercoming Data Barriers for Successful AI Projects
46mFullstack AI: The Building Blocks for Production Ready AI-Powered Software
56mPanel: Scaling AI successfully
44mPanel Discussion: The BFSI Playbook - Scaling AI with Trust and Quality
36mHow can AI turn data overload into smarter decision-making?
32mFrom Three Days to One Minute: Architecture Trade-offs in Production Enterprise AI