My Dashboard

Module summary

The My Dashboard module gives you a personalized view of your wallet’s activity and position within the network.It brings together key insights from across Hōkū — your trading history, your cluster trends, and your AI-powered token recommendations — all in one place.

To use this module, you must connect your wallet and sign in. For help on connecting, see the Getting Started page.

Hōkū uses the same machine learning models that power the 3D Wallet Explorer to identify your behavioral cluster and highlight your wallet within it. This allows you to see where you fit among other traders with similar styles, risk levels, and performance patterns.

By looking at this view, you can:

• See your position in the network. A mini 3D Wallet Explorer shows your wallet’s location within its cluster, giving quick context about your trading style relative to others.

• Track your performance. View your total portfolio balance alongside a four-week historical chart to monitor growth and volatility over time.

• Follow cluster trends. Instantly see the top trending tokens among traders who behave like you, helping you understand where your peer group is focusing their attention.

• Manage your favorite wallets. Keep an eye on the traders you follow most, all in one convenient list.

• Explore your personalized recommendations. Review your curated list of the most relevant tokens generated by Hōkū’s deep learning models, tailored to your unique trading behavior.

This feature provides a simple, high-level snapshot of everything that matters to you on-chain. It helps you understand your performance, stay aware of what similar traders are doing, and discover new token opportunities matched to your personal trading style.

Features

Network Toggle

The Network Toggle lets you switch between different blockchain networks instantly. You’ll find a simple selector that controls which network’s data you’re viewing at the top of the window.

Hōkū is currently live on Base, Ethereum, and BNB Smart Chain. When you change the network, every part of your dashboard — your portfolio, cluster position, trending tokens, and personalized recommendations — updates automatically to reflect your activity and relationships within that specific chain.

This makes it easy to monitor how your trading style and performance vary across networks, or to explore entirely different ecosystems with a single click. If you use multiple chains, switching networks ensures this module always shows accurate, real-time insights for the one you’re focused on.

All modules in Hōkū are network-specific except the Trade Signal Alerts page. That means you'll need to select a specific network to view its data on all other modules where this toggle is present

Your position in the 3D space

This tile provides a focused view of where your wallet sits within the broader trading network.

It’s powered by the same 3D Wallet Explorer technology used throughout Hōkū, showing millions of active traders plotted in three dimensions based on their behavior and style.

Each point represents a wallet, and each color represents a cluster—a group of traders who act in similar ways. For more information on the 3D visualization, see the 3D Wallet Explorer page.

In this simplified view, all controls are removed, and your wallet is automatically selected and centered, allowing you to clearly see your position within your cluster and the surrounding trader landscape. This visualization helps you understand how your trading behavior compares to others on the network.

Being close to other points means your wallet behaves similarly to those traders—such as taking similar levels of risk, trading similar token types, or maintaining comparable activity patterns.

It’s a quick, intuitive way to see where you belong in the on-chain ecosystem—whether you’re part of a large, active community of traders or positioned in a smaller, niche group that trades differently from the rest of the market.

Current and historical balances

The Current and Historical Balance tile gives you a clear snapshot of your wallet’s value and how it has changed over time.

At the top, you’ll see your current total balance, followed by an interactive chart showing your portfolio history over the past four weeks. This makes it easy to track your growth, volatility, or any large swings that occurred during specific market events.

Below the chart, your Top 5 Holdings are listed, showing the token name, current price, amount held, and total USD value. This table updates automatically as your wallet changes, so you always have an accurate view of your active positions.

You can also view your balance and holdings at specific points in the past by clicking the time buttons (Now, -7 Days, -14 Days, and -21 Days).

Each button instantly reloads your portfolio as it appeared at that time, allowing you to see how your positions evolved, for example, what tokens you held two weeks ago versus now, or how your total value has shifted over time.

This feature helps you see your trading performance at a glance and understand how your decisions have affected your portfolio value across recent weeks.

Trends from your Trading Cluster

The Trending Tokens panel shows what traders in your own behavioral cluster are doing right now.

It’s powered by the same real-time system that drives the cluster leaderboards across Hōkū but this view is focused entirely on your cluster—the group of wallets that trade most like you. For more information on our real-time cluster trends, see the Trading Clusters page.

Each token listed represents where activity within your cluster is currently concentrated. The percentage listed in each row represents the share of total transactions that token was involved in across all wallets in this cluster during the selected time period. In the below example, 5.97% of all buy transactions in the past 1 hour within Trading Cluster 3 involved the PIRATE token. The data updates continuously as new transactions occur on-chain, showing which assets traders in your group are buying, selling, or rotating into at this moment.

You can adjust filters to show specific transaction types (buy or sell), include or hide stablecoins, and choose a time window to track short-term or longer-term trends.

This feature helps you see what’s capturing the attention of traders who behave like you—whether they’re chasing new narratives, taking profits, or rotating into safer assets. It’s an easy way to spot emerging opportunities and understand your cluster’s sentiment before broader market trends become visible.

Your favorited wallets

The Favorite Wallets Preview gives you a quick look at the wallets you’ve saved to your favorites list.

Each entry shows the wallet address (in short hash form) and its current total balance, updating in real time as their portfolios change. The page is network specific, so if you've favorited wallets on different networks, you'll need to switch the network toggle to view them.

This section is designed for convenience — it provides an at-a-glance summary of the wallets you care about most without needing to open each one individually. If you want to view more details, such as their recent trades, holdings, or cluster information, you can navigate to the “My Favorites” module from the left-hand menu.

From there, you’ll have full access to each favorited wallet’s activity and performance history, making it easy to track the traders and strategies that interest you most.

Personalized token relevance scores

The Relevant Tokens tile highlights tokens that you don’t currently hold but that our models believe are most relevant to your trading behavior.

It’s powered by Deep3 Labs’ proprietary deep learning system, which was trained on more than 100 million on-chain trades. It estimates which tokens are most likely to interest you next based on what traders like you recently bought. We use the same powerful machine learning methods platforms like TikTok and Netflix use to recommend video content.

Each token is assigned a Relevance Score, shown as a percentage. This score reflects how closely your wallet’s trading history and behavioral patterns resemble those of traders who are already active and successful with that token. In other words, it helps surface opportunities that align with your unique trading profile — tokens that traders like you have previously found valuable or profitable.

Unlike traditional market scanners or popularity lists, these results are entirely personalized. Two users looking at the same market will see completely different results based on their own wallet history, cluster membership, and trading style.

This feature is part of Hōkū’s personalized recommendation system — designed to help you discover tokens that fit your strategy before they trend. It’s not about what you already own — it’s about what’s missing from your portfolio that may deserve a closer look. For more info on how we built this system, see the How it works section below.

Usage tips

This isn't just any dashboard. The Relevant tokens feature uses some of the most advanced AI in the industry that's been fully. So while we're sure you'll discover new and interesting ways to capitalize on this information, here's a few just to get you started.

Use your relevance scores to spark new trading ideas

The personalized relevance tile is designed for discovery. If you’re unsure what to research next, start with your highest-scoring tokens. These aren’t random picks — they come from patterns found in the trading activity of wallets that behave like yours. Clicking any token opens deeper information such as its performance and where it’s trending among your cluster peers. It’s a fast, data-driven way to surface opportunities that fit your trading style without spending hours digging through charts or lists.

Track how your performance compares to your cluster

The “Top tokens in your cluster” section shows what traders like you are buying right now. By comparing this to your own holdings, you can quickly see whether you’re following or diverging from your peer group. If your balance history is steady while your cluster is rotating into new sectors, that difference might signal where fresh opportunities—or risk—are forming.

FAQ

Why don’t I see tokens I already hold in my recommendations?

The system is designed to highlight new opportunities. Tokens you already hold are automatically excluded, since their relevance is already confirmed by your own activity.

How often are my recommendations updated?

Recommendations evolve continuously as your neighbor wallets' holding change. This means your list will update dynamically with the market. The underlying models are refreshed weekly, so you can expect to see the largest changes once per week.

Why are some tokens with high relevance scores unfamiliar or low-market-cap?

That’s expected. Hōkū’s models sometimes surface niche or early-stage tokens held by clusters of profitable traders similar to you. These tokens may not yet be widely traded, which can make them valuable discoveries.

What does a higher relevance score actually mean?

It means that, based on patterns learned from millions of trades, traders most similar to you are holding or buying that token more frequently than average. It’s a measure of behavioral similarity, not a financial guarantee.

Why isn’t my wallet appearing in the dashboard?

Only active wallets with recent on-chain activity are included. If your wallet hasn’t made a token transfer in the past month or has fewer than three lifetime transfers, it won’t yet appear in the system. Once you meet those criteria, it will be included automatically in the next refresh.

How often does my balance and history update?

Balances and portfolio history update as new blocks are processed. In most cases, that’s every few seconds—faster than many major blockchain explorers.

Why don’t I see any favorite wallets here?

The Favorites tile is a preview. You’ll need to mark wallets as favorites in the 3D Wallet Explorer for them to show up here. It's also network specific, so pay close attention to the network you've selected in the menu at the top of the page.

Is any of this financial advice?

No. Hōkū highlights statistical patterns and trader similarities—it doesn’t predict outcomes or guarantee profit. Always do your own research before trading.

How it works

Behind the simple tile showing your personalized token list lies one of Hōkū’s most advanced AI systems. It analyzes millions of trades from across the network to understand how different tokens appeal to different types of traders. By comparing your activity to similar wallets and adjusting for broader market trends, it estimates which tokens you’re most likely to find relevant next. This section explains how that scoring system works — from training the deep learning model on real on-chain behavior to generating your final, ranked recommendations.

It's a sophisticate system, so some complexity is unavoidable. But we've done our best to distill it all in simple terms. But, data scientists should feel right at home.

---

Step 1: Gathering the data

Every month, Hōkū collects the complete trading history of wallets that have been active during the last 30 days. This process typically includes around 5 million wallets and over 100 million individual trades across the network.

For each wallet, the system measures around 20 features, such as:

• Wallet age

• Total number of trades

• Total profit or loss

• Average time between trades

This information helps the model understand each trader’s style and activity level.

Using a rolling window helps the clusters adapt as market cycles, meta-games, and liquidity conditions change.

---

Step 2: Build wallet behavior and risk features

For each wallet we compute two kinds of signals and treat them as one combined description of behavior.

Wallet characteristics (just a few examples)

We compute over 20 different descriptive features of each wallet, which is constantly growing and evolving as our research progresses.

• Age: how long the address has been active.

• Activity: number of trades, number of distinct tokens, trading frequency.

• Capital & concentration: balances, size of typical trade, diversification.

• Performance: realized profit/loss over time, volatility of outcomes.

Contextual risk features

Past trades are also summarized by how crowded or isolated those trades were at the time:

• If a token was traded by many other wallets in the same period, that action is treated as lower risk (more consensus, more liquidity).

• If a token was traded by few wallets in that period, it is treated as higher risk (niche, less consensus, thinner liquidity).

These signals are aggregated for each wallet (for example, “share of trades in high-risk contexts” or “typical crowd size when entering positions”). Together with the characteristics above, they form a single feature vector—a compact but expressive fingerprint of that wallet’s behavior.

Combining intrinsic traits (age, frequency, outcomes) with context (how crowded their trades were) separates look-alike wallets that actually behave very differently. It separates two wallets with similar activity counts: one who follows liquid, well-known moves vs. another who hunts illiquid, contrarian bets.

---

Step 3: Learn three complementary embeddings

We represent each wallet in three ways, then later fuse them:

1. Basic behavior embedding (UMAP) – learned from wallet features like age, activity, P/L.

2. Risk embedding (SVD + UMAP) – learned from a binned matrix of wallet × token-risk exposure.

3. Graph (CF) embedding (LightGCN) – learned from the wallet–token bipartite graph built directly from counts.

Each embedding captures a different kind of intelligence:

• The basic behavior embeddings show who a wallet is (its intrinsic style).

• The risk embeddings show how a wallet behaves under pressure (its appetite for risk and liquidity).

• The graph embeddings show who a wallet is connected to (its position in the trading ecosystem).

Fusing all three creates a richer, more complete behavioral model than any single view alone.

---

Step 4: Train LightGCN on the wallet–token graph

The final and most advanced embedding comes from a Light Graph Convolutional Network (LightGCN) — the same class of model used in modern recommender systems like YouTube, TikTok, and Amazon.

Where the earlier embeddings captured what each wallet is like and how it trades, LightGCN captures who trades what and how those choices connect across the entire market.

How it works conceptually

Think of the blockchain as a vast bipartite graph linking two types of nodes:

• Wallets on one side

• Tokens on the other Each edge between them represents trading activity — a wallet that bought, sold, or transferred a given token.

LightGCN learns patterns from this network directly. Instead of manually defining features (like “number of trades” or “average profit”), it learns them by propagating signals across the graph.

Each wallet’s representation is updated based on the tokens it interacts with, and each token’s representation is updated based on the wallets that trade it. After several propagation layers, the model discovers a smooth “latent space” where:

• Wallets with similar portfolios end up close together.

• Tokens often traded by similar wallets cluster nearby.

• Rare but meaningful co-occurrences — the kind that reveal early discovery of new tokens — are preserved.

Why this approach matters

Traditional statistics or clustering methods can only look at features within each wallet. LightGCN, by contrast, captures relationships between wallets — the structure of the trading network itself.

This gives it three key advantages:

1. Collaborative context: It learns that two wallets are similar not because their metrics match, but because they act on similar opportunities.

2. Cold-start resilience: Even a wallet with limited history inherits information from the tokens it has touched, giving it meaningful placement in the network.

3. Emergent discovery: When new tokens attract a common set of early traders, LightGCN links them before price data or hype cycles make those patterns visible elsewhere.

Why we use LightGCN specifically

The “light” version of GCN omits unnecessary neural transformations between layers, keeping only the core graph propagation. That makes it both more interpretable and faster — perfect for large-scale, real-time blockchain data. Instead of millions of parameters trying to learn arbitrary shapes, it focuses on one thing: the flow of similarity across the trading network.

LightGCN captures the hidden structure of “who tends to buy what.” Rather than memorizing transactions, it learns latent affinities between traders and tokens — the same technique used by recommendation engines at TikTok and Netflix. This is the step that makes Hōkū’s AI capable of predicting which tokens you’ll find relevant next, even if you’ve never interacted with them before.

---

Step 5: Build nearest-neighbor context

Once every wallet has its fused embedding — a compact numerical fingerprint representing its trading behavior — we calculate which wallets are most similar to each other.

This is done by comparing every wallet’s vector to every other wallet’s vector in the embedding space and identifying the top-K nearest neighbors for each one.

In simpler terms, this step answers:

• Who are the traders most like this wallet, across all behavioral dimensions?

Each wallet ends up with a small group of its most similar peers — traders who share its overall profile of timing, risk tolerance, and token interests.

The nearest-neighbor table is what allows all downstream personalization. It’s the bridge between descriptive analytics (what kind of trader you are) and predictive analytics (what you might do next).

Most large-scale recommender systems use approximate methods (like FAISS or HNSW) to speed up similarity search. However, Hōkū performs exact nearest-neighbor computation via GPU matrix multiplication, ensuring that every neighbor relationship is mathematically precise.

This matters because:

• Accuracy beats approximation when wallets are highly diverse — small errors can flip similarity rankings.

• GPU linear algebra makes it practical: matrix operations that would take hours on CPU finish in minutes on modern CUDA hardware.

• Determinism ensures consistency: the same wallet will always find the same neighbors when recomputed, which is critical for reproducibility and trust in AI-driven recommendations.

This step transforms millions of scattered data points into a living behavioral network — one that evolves as traders move, tokens trend, and the market shifts. It makes all of the complex math up to this point available in a practical format that traders can act on.

---

Step 6: Score token relevance for tokens you don’t hold

To provide the final relevance scores, we compute a single weighted formula for each token your neighbor wallets hold, but you do not. It measures how relevant each token is to you by learning from the behavior of traders most similar to you. It looks at which tokens your peers hold, how many of them agree on those tokens, and how those choices compare to what’s popular across the whole network. The result is a balanced, data-driven relevance score that highlights new tokens you’re most likely to find interesting or profitable based on the collective behavior of traders like you.

Formally, its expressed as follows:

$$\operatorname{relevance}(t)= \frac{\tilde v_t-\min_{t'} \tilde v_{t'}}{\max_{t'} \tilde v_{t'}-\min_{t'} \tilde v_{t'}+\varepsilon}, \qquad \tilde v_t := \frac{k_t\!\left(\sum_{i=1}^{K} w_i\,p_{i,t}\,(1-\beta\,u_t)\right)+\lambda\,g_t}{k_t+\lambda}.$$

$$\begin{aligned} \text{where:} \quad K & : \text{number of nearest neighbors considered} \\ w_i & : \text{normalized similarity weight of neighbor } i \\ p_i[t] & : \text{neighbor } i \text{'s exposure to token } t \text{ (binary or portfolio share)} \\ u[t] & : \text{user's own holding share of token } t \\ \beta & : \text{damp factor for already-held tokens (0–1)} \\ k_t & : \text{effective peer support for token } t \text{ (weighted count of neighbors holding } t) \\ g[t] & : \text{global prior (fraction of all wallets holding token } t) \\ \lambda & : \text{Bayesian prior strength (controls influence of global popularity)} \\ \varepsilon & : \text{small constant to prevent division by zero during normalization} \\ \tilde v_t & : \text{Bayesian-shrunk weighted relevance for token } t \\ \min_{t'}, \max_{t'} & : \text{minimum and maximum taken across all candidate tokens for display scaling} \\ \operatorname{relevance}(t) & : \text{final normalized token relevance score in } [0,1] \end{aligned}$$

In essence, this is the model’s answer to one question —

“If traders who behave like me are buying something I don’t own yet, how confident should I be that I’d care about it too?”

Only tokens you don’t currently hold are scored and displayed — ensuring Hōkū’s recommendations focus on discovery, not redundancy.

This final step blends personal and collective intelligence. Your nearest peers vote on what might fit your trading style, but that signal is calibrated against global market behavior so the output stays stable and interpretable.