In an interview, Samarth Setia, Founder of Rezio.ai, speaks with TimesTech about building AI for trust-driven, offline industries. He explains how Rezio transforms scattered WhatsApp chats, calls, and documents into structured, verified intelligence. Setia shares why he chose decision support over full automation, how AI-native architecture shaped the product, and where enterprise AI delivers measurable value beyond the hype.
Read the full interview here:
TimesTech: What core technology gap did you identify in offline, trust-driven industries that led you to build Rezio AI?
Samarth: Offline industries run on conversations and relationships, but their truth data is scattered across WhatsApp, calls, PDFs, and people’s memory. That makes “what’s real right now” hard to verify at scale. Rezio turns messy conversational signals into structured, continuously refreshed intelligence—so decisions are based on verified reality, not hearsay.
TimesTech: What was the hardest technical challenge in building AI for highly fragmented, unstructured real-world data?
Samarth: The hardest part is that unstructured data isn’t just messy—it’s context-heavy and constantly changing. You’re dealing with shorthand, missing details, duplicates, outdated availability, and multiple formats (text, images, PDFs, voice). The breakthrough is combining extraction with entity resolution, confidence scoring, and verification loops so the system doesn’t just “read” data—it keeps it true.
TimesTech: Why did Rezio prioritise decision support systems over full automation, and how did that choice shape the product?
Samarth: Real estate decisions are high-stakes, emotional, and full of edge cases—full automation creates risk when nuance matters. So Rezio is designed as an intelligence layer: it shortlists, explains trade-offs, flags risks, and guides next steps, while humans retain final judgment. That choice shaped the product into an “AI broker” that improves outcomes—without pretending a model should replace responsibility.
TimesTech: How does an intent-led, WhatsApp-first interface improve data quality and system intelligence compared to traditional CRMs?
Samarth: CRMs depend on manual entry, so they capture activity—often late and incomplete. WhatsApp is where intent and inventory are exchanged in real time, so an intent-led interface captures the signal at the source. That typically reduces missing data, improves freshness, and enables verification because the system can constantly cross-check availability and details—rather than relying on static, outdated records.
TimesTech: After exiting Mr Milkman, what changed most in how you approached architecture, AI readiness, and product scale at Rezio?
Samarth: The big change was designing AI-native from day one—not “add AI later.” That means clean ingestion pipelines, strong schemas, evaluation harnesses, and feedback loops built into the core workflow. After my last journey, I’m far more deliberate about building modular systems that can scale and iterate fast without rebuilding foundations every time.
TimesTech: How do you design AI systems that work alongside human judgment instead of trying to replace it?
Samarth: We treat AI like a senior analyst: it makes recommendations, but it must show its work. So Rezio provides structured outputs, confidence levels, and clear reasoning, plus escalation to humans when uncertainty is high. Then we close the loop—human decisions and transaction outcomes feed back into the system so it improves while staying accountable.
TimesTech: What technical differences did you encounter while building the same intelligence layer for the Indian and US markets?
Samarth: When you build for India, you’re operating in a much more conversational and unstructured reality—multiple languages, WhatsApp-native workflows, fragmented documentation—so the system must excel at extraction, deduplication, and ambiguity handling, with strong verification loops. When adapting the same intelligence layer to the US, the ecosystem is generally more standardized, so integration, data governance, compliance expectations, and auditability become more central. Same intelligence goal—very different “data reality.”
TimesTech: With AI hype everywhere, where do you think enterprise AI delivers real value today—and where is it still mostly noise?
Samarth: Real value is where AI is embedded into workflows and produces measurable outcomes: faster decisions, lower operational load, higher conversion, fewer errors—especially when it converts unstructured data into usable, verified intelligence. The noise is generic copilots and flashy demos that aren’t connected to domain data, don’t change execution, and don’t improve outcomes consistently.
