[Hackathon] feat: Dataset Bank + Dataset Search Agent + Results Dashboard

[EmilySun621]

TL;DR: Browse and one-click import public datasets from UCI, Kaggle, dkNET, WHO, and PubMed — right inside Texera. Ask the AI agent to find datasets for you. Get analysis results in a dedicated dashboard instead of buried in chat.

⚠️ Still under testing

---

😤 The Problem

A researcher wants to study diabetes. Before she can even start analyzing, she has to:

1. Open a new tab, google around for datasets
2. Download a CSV from UCI, upload it to Texera, manually configure the file path
3. 20 minutes gone before any actual work
   Then after running her workflow, the AI gives her a detailed model comparison — but it's buried in a long chat message she has to scroll through. Can't reference it, can't copy it cleanly, can't share it with her advisor.

Finding data is painful. Reading results is painful. We fixed both.

---

✨ What We Built

📦 1. Dataset Bank — Browse & Import Without Leaving Texera

A new sidebar page with a curated catalog of public datasets — searchable, categorized, one-click importable.

	Feature	Details
🔎	Search	By name, description, or tag (e.g., "diabetes", "classification", "healthcare")
🏷️	Categories	Biomedical, NLP, Computer Vision, Finance, Social Science, Time Series, Tabular
🗂️	Sources	UCI, Kaggle, dkNET, WHO (Global Health Observatory), PubMed (BioNLP corpora)
📋	Rich cards	Name, source badge, description, row/column counts, file size, tags

Three actions per dataset:

Action	What it does
🔗 View on source	Opens the original dataset page for verification
⬇️ Download	Saves the file locally
☁️ Import	One click → dataset lands in "Your Datasets", ready for any workflow. No manual upload, no file path config.

Why WHO + PubMed? Biomedical researchers need more than tabular CSV data. WHO's Global Health Observatory provides structured health statistics across 194 countries — ideal for epidemiological modeling. PubMed/PMC open-access abstracts and full-text corpora are the gold standard for training biomedical NLP models (named entity recognition, relation extraction, text classification). Adding these sources means Texera's Dataset Bank covers the full spectrum from structured tabular data to unstructured text.

Backend: Server-side proxy (/api/dataset-bank/import-from-url) fetches the file and uploads it through Texera's existing dataset pipeline — bypassing browser CORS restrictions.

---

🔍 2. Dataset Search Agent Tool — "Find Me a Diabetes Dataset"

The AI agent can now search for datasets on your behalf mid-conversation.

	Feature	Details
🗣️	Natural language search	"Find me a diabetes dataset" → agent calls search_datasets tool
⚡	Parallel search	Queries dkNET, UCI, Kaggle, WHO, and PubMed simultaneously
🧠	Knows your data	Your existing Texera datasets are injected into the system prompt
🔗	Auto-configures	"Use my iris dataset" → agent knows the exact file path, sets up CSV Source automatically

---

📊 3. Results Dashboard — Analysis Reports Outside the Chat

When the AI agent produces analysis (model comparison, metrics, key findings), it now appears in a dedicated floating panel — not buried in chat.

	Feature	Details
📋	Report cards in chat	Compact card: "📊 Results ready · View Report →"
🖥️	Floating dashboard	Opens alongside the canvas — everything visible at once
📝	Formatted markdown	Tables, headers, bold metrics, key insights — properly rendered
📎	Copy & Export	Clipboard or download as markdown
🕐	Timestamped	Know when each analysis was generated
🔄	Auto-updates	New analysis from agent → dashboard refreshes

The experience: Canvas on the left (your DAG) · Chat in the middle (interaction) · Results Dashboard on the right (your analysis). Everything visible at once.

---

🎬 Demo Walkthrough

1. 📦 Open Dataset Bank → search "diabetes" → filter "Biomedical"
2. 🔗 Click "View on UCI" to verify → click ☁️ Import → dataset appears in Your Datasets
3. 💬 Ask the Diabetes Agent: "Build a classification workflow using my diabetes dataset"
4. 🤖 Agent generates the workflow on canvas
5. 🏃 Ask: "Run it and give me a comparison report"
6. 📊 "Results ready · View Report →" appears in chat
7. 🖥️ Click → Results Dashboard opens with formatted model comparison, winner, key insights
8. 📎 Copy the report to share with advisor

---

🏗️ Architecture

┌──────────────────────────────────────────────────────┐
│  🖥️  Frontend (Angular)                              │
│  • Dataset Bank page (search, categories, cards)     │
│  • Results Dashboard panel (markdown, copy, export)  │
├──────────────────────────────────────────────────────┤
│  ⚙️  Agent Service (TypeScript)                       │
│  • search_datasets tool (dkNET + UCI + Kaggle        │
│    + WHO + PubMed)                                   │
│  • User dataset injection into system prompt         │
│  • Report marker convention for dashboard            │
│  • Dataset import proxy (server-side fetch →         │
│    Texera upload pipeline, bypasses CORS)            │
├──────────────────────────────────────────────────────┤
│  🔒 Texera Core Engine (Amber) — UNMODIFIED          │
│  Everything is additive: new modules,                │
│  new endpoints, new components                       │
└──────────────────────────────────────────────────────┘

📊 +6,892 / -537 lines across 64 files · 🔒 Zero modifications to Texera's core engine

---

✅ Testing

Test	Status
Angular build	✅ Clean
agent-service typecheck	✅ Clean
Dataset import (UCI Iris, end-to-end)	✅ Pass
Results Dashboard rendering	✅ Pass
Dataset search tool (agent callable)	✅ Pass

---

💡 Why This Matters

Researchers shouldn't have to leave their analysis platform to find data, and they shouldn't have to dig through chat logs to find results.

❌ Google → download → upload → configure path → run → scroll through chat for results

✅ Search → one-click import → run → results dashboard