Building FinAI: When Instagram Becomes Your Benefits Advisor

The flow works like this: When someone enters their Instagram username, our Lambda function grabs their recent public posts. We then feed these images to Gemini's vision model with a carefully crafted prompt that looks for lifestyle indicators relevant to insurance needs. Things like physical activities, work environment, family status, and health indicators.

The genius part isn't the individual technologies - it's how we mapped visual elements to benefit recommendations. See a laptop in multiple photos? You probably need vision insurance for all that screen time. Kids showing up regularly? Life insurance becomes critical. Posting from different cities constantly? Travel insurance might make sense.

Making AI Decisions Transparent

One of our biggest breakthroughs came when we realized people don't trust black-box recommendations. Nobody wants an AI telling them what to buy without explanation. So we made every decision transparent. When the app suggests accident insurance, it shows you exactly which photos triggered that recommendation and why.

We also discovered that people respond better to scenarios than statistics. Instead of saying "accident insurance covers medical costs," we show them: "Remember that time you posted about spraining your ankle playing basketball? Without accident insurance, that ER visit would have cost $3,500. With it? Just $250."

This "what-if" approach turned abstract coverage into concrete savings. Suddenly, that $15 monthly premium doesn't seem expensive when you realize it could save you thousands.

The Privacy Problem We Had to Solve

Around 3 AM, we hit a wall. Our initial design required Instagram login credentials, and nobody on the team felt comfortable giving those to a hackathon project. If we didn't trust it, why would anyone else?

The solution was simpler than we expected. We only analyze public profiles, no login required. Users enter their username, we grab public posts, analyze them, and immediately delete the data. We even show exactly what we're looking at and let users exclude specific posts. Transparency became our privacy strategy.

For users with private profiles or those who preferred not to share their Instagram, we built a fallback: a dynamic questionnaire that adapts based on previous answers. But here's what surprised us - given the choice, 80% of our test users chose Instagram analysis over filling out forms. People hate forms that much.

The Chat Assistant That Speaks Human

We knew we needed to help users understand their benefits after selection, but nobody wants to read documentation. So we integrated a chat assistant powered by Gemini that knows your current plan and can answer questions in plain English.

The key innovation here wasn't the AI itself - it was training it to avoid insurance jargon. Ask about deductibles, and instead of a textbook definition, it responds: "Think of it like a cover charge at a club. You pay that amount first before insurance starts picking up the tab. With your current plan, that's $500 per year."

Every explanation uses analogies from daily life. Copays are explained as "like a ticket price for seeing the doctor." Networks are "doctors that agreed to give your insurance company a discount." Simple, relatable, memorable.

What 24 Hours Without Sleep Taught Us

Building FinAI in 24 hours taught us lessons that no amount of planning could have. First, your wildest idea might be your best one. When we proposed Instagram analysis at the start, even we thought it was a bit ridiculous. It ended up being the feature that won us the Technical Innovation Award.

Second, perfect is the enemy of good enough. We cut features ruthlessly throughout the night. Advanced analytics? Gone. Fancy animations? Deleted. Multi-platform support? Maybe later. What remained was a focused app that did one thing extraordinarily well - made benefit selection actually enjoyable.

Third, real data beats assumptions every time. We assumed people would be concerned about privacy with Instagram analysis. In reality, they were more concerned about the time it takes to fill out forms. We thought young people would want minimal coverage to save money. Actually, once they understood the risks through our what-if scenarios, they selected more comprehensive coverage.

The Results That Surprised Everyone

After our five-minute presentation (running on zero sleep and pure adrenaline), the judges were skeptical but intrigued. Then we showed them the numbers from our overnight testing. Users completed the entire benefit selection process in under 4 minutes, compared to the industry average of 18 minutes. Engagement was through the roof - people actually enjoyed the process.

But the real validation came from Lincoln Financial's executives. They saw something we had built as a hack turning into a genuine solution to a problem they'd been trying to solve for years: making benefits accessible to younger employees. The Technical Innovation Award was nice, but knowing we'd built something that could actually help people? That made the missed bus connections and sleepless night worth it.

Why This Approach Actually Works

The success of FinAI comes down to meeting users where they are, not where we think they should be. Traditional benefit enrollment expects people to understand complex financial products and make decisions about hypothetical future scenarios. We flipped that model completely.

Instead of asking users to imagine their future health needs, we analyze their current lifestyle. Instead of explaining what benefits cover, we show what they would save in real situations. Instead of forcing them through long forms, we use data they've already created.

This isn't about making insurance "cool" or "fun" - it's about making it relevant. When someone sees their actual lifestyle reflected in their benefit recommendations, when the amounts are explained in terms of their daily coffee budget, when scenarios reference their actual activities, insurance stops being abstract and becomes personal.

What's Next for FinAI

Winning the hackathon was just the beginning. We're now working on expanding beyond Instagram to include other lifestyle indicators. Spotify listening habits could indicate stress levels. Strava runs show fitness commitment. LinkedIn shows career trajectory and income potential.

We're also exploring how this approach could work for other complex financial decisions. Retirement planning based on spending patterns. Investment recommendations based on life goals visible in social media. The principle remains the same: use the digital footprint people naturally create to simplify complex decisions.

The insurance industry is ripe for this kind of innovation. While companies spend millions on marketing to make insurance "relatable," we showed that the solution might be simpler: actually relate it to people's real lives.

The Real Innovation

Looking back, the technical implementation wasn't the innovation - anyone can call an API or build a Flutter app. The innovation was recognizing that the data needed for personalized benefit recommendations already exists in the photos people post every day. We just had to be brave enough (or sleep-deprived enough) to connect those dots.

Sometimes the best solutions come from questioning assumptions everyone takes for granted. Who says insurance forms need to be forms? Who decided benefit selection has to be boring? Why can't an Instagram post be more informative than a health questionnaire?

That's what hackathons are really about. Not just coding for 24 hours straight, but having the freedom to try ideas that seem too crazy for the real world. Sometimes, those crazy ideas are exactly what the real world needs.

Built at codeLinc 10 Hackathon in Radnor, Pennsylvania
Technical Innovation Award Winner sponsored by Cognizant

Tech Stack: Flutter, AWS Lambda, AWS Cognito, Google Gemini API, Python, Dart

Team: Sakshi Patel, Trisha Gite, Lucas Walton, and me