Sentio

Inspiration

The spark for Sentio came from a conversation with a wealth advisor managing dozens of high-net-worth clients. She told us about her mornings: arriving at 6 AM to scan Bloomberg, WSJ, Financial Times, and industry newsletters before client calls. "I spend more time reading news than talking to clients," she said. "And I still miss things."

We dug deeper. Wealth advisors managing diverse portfolios face an impossible task. Each client is unique—an agritech investor needs agricultural policy updates, a tech executive watches IPO markets, a real estate magnate tracks interest rates. Traditional keyword alerts don't understand context. They match "agriculture" but don't know why federal subsidy changes matter to someone who owns farmland and invests in agritech startups. The false positives are overwhelming. The misses are costly.

Meanwhile, explaining multi-asset portfolios to clients meant static spreadsheets or clunky legacy software that looked like it was built in 2005. High-net-worth clients expect better. They deserve better.

We realized AI could solve this—not by replacing advisors, but by giving them superpowers. What if AI could understand context like a human? What if building a portfolio visualization was as intuitive as moving sticky notes on a board? That's when Sentio was born: an intelligent operating system that makes every advisor feel like they have a team of analysts working for them 24/7.

What it does

Sentio is built on three pillars that transform wealth management from reactive to proactive.

First, our AI-powered news intelligence layer continuously monitors markets. Every morning, we fetch the latest news from premium sources and analyze each article against individual client profiles. GPT-4 doesn't just match keywords—it understands why a federal law change matters to an agritech investor but not a tech executive. It catches indirect relevance that traditional systems miss. Each article gets a 0-100 relevance score, and only articles scoring above 50 make it through. The result? Advisors get one beautiful daily digest email summarizing all relevant news across their entire client base, with AI-generated insights explaining exactly why each article matters. What used to take 10 hours now takes 10 minutes.

Second, we built an interactive portfolio dashboard that makes complex wealth management feel simple. Using a drag-and-drop canvas, advisors create visual representations of client portfolios with widgets for stocks, bonds, real estate, private equity, crypto, and alternative investments. Want to show how a client's real estate holdings connect to their private equity stakes in construction companies? Just draw a line. Each dashboard is saved per-client and features our signature "old money" design aesthetic—navy, gold, and cream with glass-morphism effects that feel premium without being flashy.

Third, everything lives in a centralized client management system. Advisors configure news preferences for each client—keywords, categories, priority thresholds, excluded topics. The system learns and adapts. As portfolios evolve, so do the insights. It's a living, breathing command center for wealth management.

How we built it

We architected Sentio as a modern full-stack application with AI at its core. The frontend is built with React 18 and React Flow, giving us a powerful canvas-based interface for drag-and-drop dashboards. We chose Tailwind CSS for rapid styling and created a custom "old money" color palette that screams sophistication. React Router v6 handles navigation between client pages seamlessly.

The backend runs on Node.js with Express, serving a REST API that manages clients, dashboards, and news alerts. We use Supabase and PostgreSQL for data persistence, storing everything from client profiles to complex React Flow layouts as JSONB. This gives us the flexibility of NoSQL with the power of relational databases. For email delivery, we integrated Gmail API with OAuth2, ensuring our digest emails land in advisor inboxes reliably.

The real magic happens in our AI pipeline. We built a five-step analysis system: pre-filtering articles by category and priority, batch-analyzing up to 15 articles per client using GPT-4, scoring each article's relevance from 0-100, generating personalized summaries that explain why each article matters, and finally synthesizing everything into a coherent advisor digest. We carefully engineered our prompts to include just enough client context—occupation, net worth, age, keywords, risk tolerance—without hitting token limits. The result is intelligence that feels human.

We integrated four external APIs—Tavily for news aggregation, OpenAI for analysis, Gmail for delivery, and Supabase for storage—and made them work together harmoniously. Every night, our system wakes up, fetches news, analyzes it against every client profile, and delivers insights before advisors start their day.

Challenges we ran into

Building Sentio pushed us to our absolute limits, and honestly, that's what made it incredible.

Three of us on the team had just learned to code. We weren't coming from Stanford CS programs or five years at Google. We learned React by building this project. We learned about API authentication by breaking Gmail OAuth twelve times before figuring it out. When we hit walls—and we hit so many walls—we had to teach ourselves the solutions. There were moments at 2 AM where we questioned if we could actually pull this off. But every breakthrough felt like winning a championship.

The AI relevance scoring was our first major crisis. We set the threshold at 60 out of 100, thinking we'd only deliver the highest-quality matches. What happened? Almost nothing got through. Advisors would get empty digests. We panicked. Were we fundamentally wrong about the approach? After hours of debugging, we realized AI needs to be tuned like a musical instrument. We lowered the threshold to 50, made the pre-filtering more lenient, and let GPT-4 do what it does best—understand nuance. Suddenly, magic. Articles started matching beautifully. The lesson? Start lenient, then optimize. Don't strangle your AI before it can breathe.

Gmail OAuth nearly broke us. Tokens would expire, our emails would stop sending, and we couldn't figure out why. For people who'd never dealt with OAuth before, it felt like black magic. We read documentation until our eyes hurt. We rebuilt the authentication flow three times. We created detailed setup guides so no one else would suffer like we did. When we finally got that first automated digest email in our inbox, we literally cheered. It was 3 AM. We didn't care.

The database schema kept evolving as we added features—alert configs, dashboards, news alerts. Managing migrations on a remote Supabase database while three people are coding simultaneously? Chaos. We'd accidentally overwrite each other's changes. Tables would have duplicate columns. We learned the hard way to organize migrations, document everything, and make scripts idempotent so they could be run multiple times safely. By the end, we had 14 SQL files properly organized with a README explaining execution order. That discipline saved us.

React Flow performance taught us that beautiful UIs require optimization. With dozens of widgets on screen, the canvas would lag. Edge updates felt sluggish. We learned about React hooks, memoization, and batched state updates. We went from computer science theory to practical engineering in real-time. Every performance improvement felt like leveling up our skills.

But the biggest challenge? Believing we could do it. When you're new to coding and you're trying to build something this ambitious, imposter syndrome hits hard. Every error message feels personal. Every bug feels like proof you don't belong. We pushed through because we believed in the vision. And now, seeing it work? Seeing real AI analysis, real dashboards, real emails going out? That's the feeling that makes everything worth it.

Accomplishments that we're proud of

We built something real. Not a prototype, not a toy—a production-ready AI system that actually works. Our AI pipeline fetches news, analyzes it with GPT-4, generates summaries, and delivers actionable insights. It handles rate limits, errors, and edge cases gracefully. That's not trivial, especially for people who learned React six months ago.

The UI we created doesn't look like a student project. It looks professional. The "old money" aesthetic with navy, gold, and cream colors combined with glass-morphism effects creates an experience that high-net-worth clients would actually want to use. The drag-and-drop dashboard builder is genuinely delightful. We're proud of that.

But more than the technology, we solved a real problem. Advisors really do spend 10-15 hours per week on manual news scanning. Sentio reduces that to minutes. That's time they get back to spend with clients, to build relationships, to do the human work that AI can't replace. We didn't build technology for technology's sake. We built something that matters.

We also proved something to ourselves. Three team members who just learned to code built a full-stack AI application with multiple API integrations, OAuth flows, database migrations, and production-ready features. We organized 35+ documentation files, wrote comprehensive READMEs, and created a codebase that's ready for team collaboration. We didn't just write code—we became engineers.

And finally, we shipped. In a world where most projects die in idea stage, we actually finished. That's the accomplishment we're most proud of.

What we learned

AI is best used as a filter, not a generator. We discovered that GPT-4 excels at analyzing existing content—scoring news articles, identifying relevance, spotting patterns—but it needs careful prompt engineering to generate quality summaries. The key was giving it just enough context. Too little and the analysis is generic. Too much and you hit token limits and burn money. Finding that balance taught us that AI is a tool that requires craftsmanship.

User experience trumps technical complexity. We could have built a complex rules engine with Boolean logic and weighted keywords for news matching. But the AI-powered approach is simpler and works better. Users don't care about the technology—they care that it works. The drag-and-drop dashboard feels magical precisely because it hides the complexity of React Flow and JSONB storage. Good engineering is invisible.

Documentation is a feature, not an afterthought. We spent real time organizing documentation into logical categories, writing clear setup guides, and creating migration READMEs. That investment paid off every time we onboarded someone new or revisited code we'd written weeks ago. Good documentation is good developer experience.

Start lenient, then optimize. Our AI scoring started too strict, filtering out too much. By beginning lenient and observing real results, we found the right balance. Premature optimization kills projects, especially AI projects where behavior is hard to predict.

OAuth is hard, but the lessons are worth it. Implementing Gmail OAuth was painful. But it taught us about secure authentication, token refresh strategies, error handling, and user flows. Those are fundamental skills that transfer to every future project.

And perhaps most importantly: you don't need years of experience to build something amazing. You need curiosity, persistence, and a willingness to learn by doing. Three of us learned to code by building Sentio. We learned React, Node.js, SQL, API integration, OAuth, AI prompting, and deployment. The best education is building something real.

What's next for Sentio

The immediate future is about refinement and scale. We're building a client-facing portal so that advisors can share curated insights directly with clients. We're enhancing the AI with sentiment analysis to score articles as positive, negative, or neutral for portfolios, and adding impact prediction to estimate how news might affect portfolio values. Real-time data integration is next—connecting to financial APIs for live stock prices, bond yields, and real estate valuations in dashboard widgets.

Long-term, we see Sentio becoming proactive, not just reactive. Imagine the system telling an advisor, "Based on today's Fed announcement, consider rebalancing Client X's bond portfolio," or "Client Y's real estate holdings overlap with areas affected by new zoning laws." That's where AI truly becomes a co-pilot.

We're also thinking about compliance and audit trails. Wealth management is heavily regulated, and having automated documentation of advisor actions and client communications could be game-changing. Add mobile apps for on-the-go access, a marketplace for third-party widgets and integrations, and eventually a white-label solution that wealth management firms can brand and deploy—Sentio could become the operating system that runs every advisor's practice.

The vision is ambitious. But if we learned anything from building this, it's that ambitious visions paired with relentless execution create something special. Sentio started as a question: What if AI could help advisors be superhuman? Now it's a reality. And this is just the beginning.