moneyheap

TL;DR: moneyheap is a vibecoded AI investing agent built to replace hype with clarity: it monitors markets + news, filters the noise, explains what changed/why it matters/what’s priced in, and outputs clear Bullish/Bearish/Neutral views across timeframes - while staying disciplined about trust, latency/cost, and “not financial advice” constraints.

👉 The ~200-word Gemini Integration write-up is included at the very end of the document under:
“Brief text write-up (suggested ~200 words) that describes the Gemini Integration”.

Inspiration

We are living in a world where the traditional path to wealth feels hard every year. Too many people are pushed into casino-like decisions - memes, gurus, hype - when what they actually need is a disciplined process.

People want agency over their financial future, and the easiest thing to sell them is hope. We want to sell clarity: an AI service that gives you a structured view of what’s happening (news, fundamentals, flows, sentiment, technical analysis) and what it implies - so users can act with confidence without needing moonshots.

The real problem in our world isn’t lack of opportunity - it’s, actually, decision overload. Data is everywhere, interpretation is scarce. We’re building moneyheap to compress the mess into a few high-quality insights: what changed, why it matters, what’s priced in, and what risks you’re taking.

What it does

moneyheap is an AI investment agent that continuously monitors multiple data streams and turns them into a small number of clear updates.

It combines real-time market data with verified news sources and uses Gemini to:

Filter the noise: It reads thousands of headlines and identifies only high-impact events.
Analyze the trend: It processes stats like price history, volume, etc. and sentiment to generate a concise analysis.
Predict the move: It outputs a specific Bullish, Bearish, or Neutral sentiment for three distinct timeframes (7 days, 1–3 months, and 6 months), complete with a confidence score and reasoning layer.

By doing so, moneyheap gives an overview of the market, gives one daily pick as the stock/ETF with the most potential, highlights the most impactful news, and makes it easy to customize your own experience with a personal dashboard, email alerts, and smart search over the market.

We treated the system as a loop that repeatedly answers four questions:

What changed? (events + movement detection across sources)
What does it imply? (contextual interpretation, not just summarization)
What’s priced in vs. surprising? (separating noise from signal)
What should the user pay attention to? (personalization + prioritization)

A useful mental model is: data → signals → reasoning → brief.

One way to think about it is minimizing the user’s uncertainty after each update:

$$ [ \text{Decision Clarity} \propto \frac{\text{Relevant Signal}}{\text{Noise} + \text{Cognitive Load}} ] $$

Our product goal is to increase “Decision Clarity” by compressing high-entropy inputs into a few stable, explainable outputs.

How we built it

We are a team of 3 people who discussed the idea of moneyheap for several months. We all have been really into the stock market and investing for a couple of years, and we didn't see any market research tool built by people for people. AI nowadays is a fascinating game changer in everything, so we saw an opportunity and decided to build the thing. A large part of that came from this: we ran into a hackathon and tried Google AI Studio for the first time.

At first we did some experiments with UI - Gemini blew us away. Quite soon we already developed backend in Firebase and started to focus on UX instead of code - it quickly became a habit and de-facto way of work.

A bit later we bought a domain - moneyheap.ai - because we decided that we will continue with the project further even after hackathon.

The tool we’ve just built has already become our only market research routine. We track our tickers, wait for alerts, and adjust our live investments accordingly.

Challenges we ran into

TL;DR: We’re trying to build something actually useful without paying a fortune for data, waiting forever for deep analysis, or accidentally sounding like we’re giving financial advice - and on top of that we’re fighting LLM hiccups and AI Studio being laggy sometimes

The main challenges we have:

Data scarcity / limits: real-time and premium market data is expensive. We had to design around constraints and still deliver value, especially at our early stage right now.
Latency vs depth trade-offs: “Always-on” monitoring conflicts with deep analysis - but both are valuable. The system needed to handle quick passes for detection, deeper passes for the most important items, and caching/reuse where possible. The main question is: How to analyze the whole giant void that market is with LLMs for reasonable time and cost?
User trust: finance is high-stakes. We had to avoid hype and focus on transparency and repeatability. It is hard because of inner urgency to recommend something.
Signal vs noise: markets generate constant “breaking” content. The hardest part is deciding when something is truly material versus just loud.
Regulations and laws: it is hard to find the right language of financial service - nothing should sound and look like recommendation.
Gemini limits: sometimes the model is overloaded and unavailable, probably because of free tier though.
AI Studio lags: there are moments when AI Studio becomes extremely slow/laggy and eats up a lot of memory. It makes it hard to work in those moments.

Accomplishments that we’re proud of

We are proud that:

we started with moneyheap!
we’ve actually built a product (ok, MVP)
our product turns out to be useful at least for us (we already invest with its help)
we’ve built front-end, back-end, DBs and all is running 24/7 smoothly with scheduled tasks, math, external APIs, caching
we are cost-effective - and we have more tricks in mind
we bought moneyheap.ai domain and connected it to the service

And finally, we are proud that we are building and having fun.

What we learned

Vibe coding is great, it works, it is fantastic
AI Studio combines power of Gemini, its knowledge over other Google services (GCP, Firebase) and UI preview - insane combination for UX-first software development
To build a better product we need money (probably we will look for investments)
Financial sector is highly regulated and requires a lot of attention to the language your app speaks. No buy/sell - only bullish/bearish; no “we recommend” - only “market sentiment shows,” etc.
We learned a lot about many Google services like Firebase that we’ve never tried before

What’s next for moneyheap

We’re doubling down on a product that users can rely on daily.

Our focus points:

Finding investments
Dive deep into laws and regulations around such services
Making paid plans
First users, traction
Custdev and product market-fit
Personalisation per user
More agency for our engines, more data sources, more stats. Make a full blown research agent.
...
profit

Brief text write-up (suggested ~200 words) that describes the Gemini Integration

Reasoning: Moneyheap uses Gemini 3 Pro as the core reasoning layer that turns market inputs (like market stats, news) into a small number of clear, structured insights. Gemini 3 Pro sits behind every daily brief and dashboard card: it ingests curated news bundles, price/volume snapshots, technical indicators, and lightweight sentiment signals, then reasons on all that and explains what changed, what’s likely already priced in, and what risks could flip the story, without drowning users in charts.

Function calling: We rely on Gemini’s tool (function) calling to search for the news and data in agentic mode. Here we use Gemini 3 Flash to optimize for speed, low latency, and high-volume tasks. Because markets are time-sensitive, we ground important summaries using provided source excerpts (and optionally Google Search grounding), and we use Gemini context caching to reuse schemas/watchlists for lower latency and cost.

Structured JSON Generation: The application relies heavily on responseSchema to enforce strict JSON outputs. Instead of returning generic chat text, Gemini returns structured data objects containing specific fields for "Short/Mid/Long Term" prognoses, confidence scores (0-100%), and sentiment labels.This allows the frontend to render complex AI reasoning into clean UI components like gauges, badges, and comparison tables.