“I have always found insurance as a sector that’s disorganised and lacks clarity for common people to understand. Is it something that AI can streamline?” 20-year-old Advith Sharma decided upon making it happen with his startup, Aether Labs, which used AI to streamline the entire insurance infrastructure.
Across India, millions of insurance agents currently run their daily operations using traditional, “jugaad” setups like Excel sheets and WhatsApp text messages. The operational data in this industry is rarely standardised, forcing agents to constantly cycle between spreadsheets, PDFs, WhatsApp conversations, email attachments, and isolated legacy insurer portals – all this just to service their clients.
Advith’s Aether Labs now wants to address this chaos with Flow AI. It is a smart platform designed to transform fragmented operational data from insurance into a clean, automated operating system for the agencies.
The platform automates complex insurance workflows like policy issuance, renewals, and endorsements, turning tasks that traditionally took days into seamless processes completed in minutes.
Advith’s practical approach to building software
What makes Advith’s journey remarkable is that he stepped into the highly technical world of AI orchestration with absolutely no prior coding experience and without a technical co-founder. Building robust AI-powered workflows and natural language model interfaces is notoriously complex, but his time as a student at the Scaler School of Technology played a meaningful role in his rapid upskilling.
The environment there encouraged thinking, rapid experimentation, and learning by building rather than learning purely through theory. This mindset helped Advith break down complex technical systems into smaller problems and solve them one step at a time. Instead of following a rigid theoretical path, he learned by shipping real code, spending months deep-diving into product design, software architecture, APIs, databases, workflow systems, and AI tooling while simultaneously building live products. Every problem encountered forced him to understand another layer of the stack.
“The advantage of starting in the AI era is that learning cycles are dramatically shorter than they used to be,” Advith explains. “When I started, I wasn’t approaching the problem as someone trying to become a traditional software engineer. I approached it as a founder obsessed with solving a specific industry problem. That changes how you learn.”
Advith notes that vibe coding tools have completely democratised software creation, lowering the barrier to entry for builders. “Today, a founder with enough curiosity, persistence, and product understanding can build far more than was possible even a few years ago,” he says.
Ultimately, Aether Labs was built through relentless iteration, learning in public, and solving real customer problems one workflow at a time.
De-complexing insurance
The core challenge that Flow AI attacks is the complete lack of standardisation across legacy insurance providers. Every individual company uses entirely different layouts, formats, and industry terminology, making data entry a major challenge.
Advith built Flow AI as an orchestration layer rather than a standalone AI model. It combines state-of-the-art foundational models with proprietary insurance workflows, structured data pipelines, and domain-specific reasoning systems. It leverages leading LLMs for understanding natural language, document extraction, and workflow execution.
On top of these models, Aether Labs has built its own insurance intelligence layer that natively understands policy structures, insurer-specific formats, servicing workflows, and compliance requirements.
“Most of the value isn’t in the base model itself, it’s in the workflow architecture around it,” Advith points out. “We’ve spent significant time building systems that can connect documents, customer records, renewal schedules, insurer requirements, and servicing actions into a single automated flow. This allows agents to automate tasks that traditionally required multiple people, spreadsheets, and follow-ups.”
To process varied policy PDFs from dozens of different legacy providers, the platform combines OCR, document intelligence pipelines, layout understanding, entity extraction, and large language models to process both structured and semi-structured documents. To prevent errors, Flow AI applies strict confidence scoring and validation mechanisms to detect inconsistencies.
“This is important because even small extraction errors can create operational issues later in the policy lifecycle,” Advith adds. “The objective isn’t simply reading PDFs, it’s converting documents into reliable, actionable data that can drive automation.”
Once the information is structured, renewals can be tracked automatically, and servicing requests can be generated proactively without requiring manual data entry.
‘Insurance GPT’ for helping with decisions
Then there’s a consumer-centric Insurance GPT, which recommends insurance policies based on specific user needs. Because insurance is a highly regulated, high-stakes financial sector, allowing an AI to casually “guess” or hallucinate details will be an absolute dealbreaker.
Hence, Advith spent months making the AI learn the dense nuances of policy wordings, underwriting principles, IRDAI regulations, product structures, exclusions, and claim processes. Instead of training the AI to generate responses from memory, Insurance GPT is designed tightly around rigid retrieval and verification systems.
“When a user asks a question, the AI references structured product data, policy documents, insurer information, and predefined decision frameworks before generating a response,” says Advith. “This significantly reduces the risk of hallucinations because the system is grounding its answers in verified information,” he adds.
Crucially, the tool is explicitly positioned as an educational and recommendation layer rather than a replacement for regulated financial advice. The system explains why it is recommending a policy, what assumptions it is making, and exactly where users should verify details directly with insurers or licensed advisors.
As Advith highlights, “In high-stakes financial decisions, transparency and explainability matter just as much as intelligence.”
Bringing AI to the grassroots level
“The biggest UX mistake AI companies make is forcing users to learn new behavior,” Advith asserts. “Most insurance agents don’t wake up wanting to use AI. They want to issue policies faster, manage renewals, and serve customers efficiently.”
By focusing heavily on plain-language interactions and intuitive one-click workflows, the software bypasses complex technical jargon entirely. “A good benchmark for us is WhatsApp,” notes Advith. “If a feature requires extensive training, it’s probably too complicated. The best AI experiences disappear into the workflow and feel like a natural extension of what users already do. Our philosophy is that AI should reduce cognitive load, not increase it,” he added.
Will AI replace humans in insurance?
“We view AI as a force multiplier for agents, not a replacement for them. Insurance is fundamentally a trust-driven industry,” assures Sharma. “Customers often make decisions involving their family’s health, financial security, or business risks. Those conversations require empathy, judgment, and context that humans provide exceptionally well. What AI can do is eliminate repetitive work,” he adds.
By allowing AI to handle routine questions, prepare policy comparisons, organise documents, and track renewals instantly, agents are freed from back-office administrative burdens. This allows them to spend high-value time advising customers and building relationships – just as most world leaders and CEOs of AI companies keep suggesting.
“The future we see is not humans versus AI. It’s human expertise amplified by AI,” concludes Sharma. “The most successful agents will be the ones who leverage AI as a co-pilot while continuing to provide the trust and guidance that customers value.”
Editorial Note: This profile is based on original reporting, including direct communication with Advith Sharma. To ensure a comprehensive perspective, FinancialExpress.com corroborated this information with public records and third-party sources. This content is not sponsored, and FinancialExpress.com retains full editorial independence and final authority over all editorial decisions.
