Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/holzerjm/civichacks-demo/llms.txt

Use this file to discover all available pages before exploring further.

The CityHack dataset (cityhack_boston_311.txt) is a synthetic 311 service request analysis for Boston, 2025. It contains detailed metrics on service volumes, resolution times, geographic equity gaps, staffing challenges, language access, and open issues.

​
Dataset overview

File: data/cityhack_boston_311.txt
Scope: City of Boston 311 Service Request Analysis, 2025
Size: ~31 lines, 13,800+ characters
Format: Plain text, structured report

​
Data sections

​
System-wide metrics

  • 487,293 service requests processed in 2025 (+12% YoY)
  • Average resolution time: 8.4 days (up from 7.1 days in 2024)
  • Customer satisfaction: 3.2/5 stars (down from 3.5 in 2024)
Request channels:
  • Phone: 38%
  • Mobile app: 34%
  • Web portal: 22%
  • Twitter/social media: 6%

​
Sample queries

Here are the three built-in queries for the CityHack track: 
python scripts/demo_step2_rag.py city 1

​
Query 1: Which neighborhoods have the longest 311 response times and what are the equity implications?

This query surfaces:
  • Mattapan: 14.1 days (vs. 4.2 in Back Bay)
  • Roxbury: 12.4 days
  • Dorchester: 11.8 days
  • Income correlation: 2.8x longer times in <$45K income areas
  • Satisfaction scores: 2.4-2.9 vs. 4.1 in high-income areas

​
Query 2: What are the biggest service gaps for non-English speaking residents?

This query identifies:
  • 32% longer resolution times for non-English requests
  • Translation only available for phone (not app/web)
  • 18% lower resolution rate in non-English areas
  • Spanish (8.2%), Haitian Creole (4.1%), Chinese (2.8%) top languages
  • Unfunded multilingual chatbot request

​
Query 3: What patterns suggest systemic inequity in city service delivery?

This query highlights:
  • 2.8x resolution time gap by income
  • Mattapan street light outages: 21+ days (vs. citywide 14.8)
  • 45% of street cleaning requests from 3 neighborhoods (Dorchester, Roxbury, Allston-Brighton)
  • Non-English areas: 18% lower resolution rate
  • 22% call abandonment rate (up from 15%)

​
Key metrics reference

MetricHigh-income areasLow-income areasMultiplier
Avg resolution4.2-5.8 days11.8-14.1 days2.8x
Customer satisfaction3.8-4.1/52.4-2.9/5-31% to -42%
Income threshold>$100K median<$45K median—
Category2025 requestsAvg resolutionNotes
Street Cleaning67,4213.2 daysPeak: Apr-Jun
Pothole Repair52,38811.7 days23% duplicates
Code Enforcement48,10222.3 daysStudent move spikes
Street Lights41,75614.8 days15% repeat reports
Tree Maintenance38,29145.2 days6+ month routine backlog
Language% of requestsResolution time impact
English85.6%Baseline
Spanish8.2%+32%
Haitian Creole4.1%+32%
Chinese2.8%+32%
Vietnamese1.9%+32%
Portuguese1.4%+32%
Gap: Translation services unavailable on mobile app and web portal (56% of requests).
Metric20242025Change
Full-time agents—42—
Part-time agents—15—
Avg wait time4.2 min7.8 min+86%
Abandonment rate15%22%+47%
Budget—$4.8M-3% (inflation-adjusted)

​
Using this dataset in the web app

When you select CityHack in the Gradio app (Step 3), the interface displays:
  • Header: “🏙️ CityHack — Analyze Boston 311 service requests, equity gaps, and service delivery patterns”
  • Example questions: All three queries above as clickable examples
  • Chat responses: AI answers grounded in the specific metrics from this dataset
Try asking follow-up questions like “Why does Mattapan have such long wait times?” or “What’s causing the code enforcement spikes?” to explore the data interactively.

​
Querying from code

Here’s how the RAG pipeline loads and queries this dataset: 
from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, Settings
from llama_index.llms.ollama import Ollama
from llama_index.embeddings.huggingface import HuggingFaceEmbedding

# Configure local AI
Settings.llm = Ollama(model="llama3.1", request_timeout=120.0)
Settings.embed_model = HuggingFaceEmbedding(model_name="all-MiniLM-L6-v2")

# Load the CityHack dataset
data_file = "data/cityhack_boston_311.txt"
documents = SimpleDirectoryReader(input_files=[data_file]).load_data()

# Build vector index
index = VectorStoreIndex.from_documents(documents)
query_engine = index.as_query_engine(streaming=True, similarity_top_k=3)

# Query
response = query_engine.query(
    "Which neighborhoods have the longest 311 response times and what are the equity implications?"
)
response.print_response_stream()
See Step 2: RAG with civic data for the full implementation.

Build docs developers (and LLMs) love

Get started for freeTalk to us