Powder

An AI agent that answers: "Where should I ski/snowboard today?"

Given a natural language query, Powder recommends the best Northeast US mountain based on current conditions, your location, pass type, and preferences.

Features

Query Understanding

The agent parses natural language queries to extract:

Filter Type	Examples
Pass type	"I have an Ikon pass", "Epic pass holder"
Drive time	"within 2 hours", "max 3 hour drive"
Terrain parks	"looking for a good park", "rails and jumps"
Glades/trees	"tree skiing", "want to ski glades"
Beginner terrain	"teaching my kid", "first time skiing"
Expert terrain	"double blacks", "steep chutes"
Night skiing	"after work", "night skiing tonight"
Skill level	"intermediate snowboarder", "expert skier"

Mountains Covered

31 Northeast US mountains with full metadata:

Vermont (9): Jay Peak, Killington, Mad River Glen, Mount Snow, Okemo, Smugglers' Notch, Stowe, Stratton, Sugarbush
New Hampshire (9): Attitash, Bretton Woods, Cannon Mountain, Cranmore, Gunstock, Loon Mountain, Mount Sunapee, Waterville Valley, Wildcat Mountain
New York (6): Belleayre Mountain, Gore Mountain, Holiday Valley, Hunter Mountain, Whiteface, Windham Mountain
Massachusetts (4): Berkshire East, Jiminy Peak, Nashoba Valley, Wachusett Mountain
Maine (3): Saddleback, Sugarloaf, Sunday River

Each mountain includes: coordinates, vertical drop, trail counts, terrain percentages, terrain parks, glades, pass types (Epic/Ikon/Indy), lift types, snowmaking %, learning facilities, pricing.

Example Schema (Killington):

{
"name": "Killington",
"state": "VT",
"lat": 43.6045,
"lon": -72.8201,
"vertical_drop": 3050,
"num_trails": 155,
"num_lifts": 22,
"green_pct": 17,
"blue_pct": 40,
"black_pct": 33,
"double_black_pct": 10,
"terrain_parks": "easy,intermediate,hard,superpipe",
"glades": "easy,intermediate,hard",
"pass_types": "ikon",
"allows_snowboarding": true,
"lift_types": "gondola,bubble,highspeed,fixed",
"has_night_skiing": true,
"avg_weekday_price": 145,
"avg_weekend_price": 190,
"snowmaking_pct": 71,
"has_magic_carpet": true,
"has_ski_school": true,
"learning_area_quality": "excellent"
}

Data Sources

The agent combines three types of data:

Context (Provided at Query Time)

Input	Description	Example
`query`	Natural language request	"Best powder with Ikon pass?"
`location`	User's starting point	Boston (lat: 42.36, lon: -71.06)
`date`	Target ski date	2025-02-17

Database (Static Mountain Metadata)

Queried via search_mountains() tool - see schema above. Filters by:

Pass type (Epic/Ikon/Indy)
Terrain features (parks, glades, night skiing)
Geographic bounds

Tools (Dynamic API Calls)

Tool	Source	Returns
`get_conditions()`	Open-Meteo API	Fresh snow (24h), snow depth, temp, wind, visibility, weather code
`get_drive_time()`	OpenRouteService API	Duration (minutes), distance (miles)
`check_crowd_level()`	Date logic	Holiday/vacation week detection (Christmas, MLK, February break)

Architecture

Two agent implementations with tools available to both.

Tools:

Tool	Args	Returns	Notes
`search_mountains`	max_drive_hours, pass_type, needs_terrain_parks, needs_glades, needs_night_skiing, needs_beginner_terrain, needs_expert_terrain, user_lat, user_lon	JSON list of mountains	Uses Haversine prefilter. Agent must pass location explicitly.
`get_mountain_conditions`	lat, lon, target_date	JSON with temp, wind, snow depth, fresh snow, weather	Agent calls per-mountain (no batch).
`get_driving_time`	start_lat, start_lon, end_lat, end_lon	JSON with duration_minutes, distance	Agent must pass both origin and destination coords.
`check_crowd_level`	target_date, mountain_state	JSON with crowd_level, vacation_week, crowd_note	Holiday/vacation detection.

1. Pipeline (Explicit Flow)

Fixed 4-step flow with embedded tool calls and intermediate outputs at each stage:

flowchart LR
subgraph Context
Q[Query]
D[Date]
L[Location]
end
subgraph Tools
C[get_mountain_conditions]
DT[get_driving_time]
CR[check_crowd_level]
S[(search_mountains)]
end
subgraph Signatures
P[ParseSkiQuery]
A[AssessConditions]
SC[ScoreMountain]
G[GenerateRec]
end
Q --> P --> S
L --> S
S --> C
S --> DT
D --> C
D --> CR
L --> DT
C --> A
DT --> A
A --> SC
SC --> G
CR --> G
G --> R[Recommendation]

DSPy Signatures (LLM calls):

Signature	Input	Output	Purpose
ParseSkiQuery	Query + user_context	Structured filters	Extract pass type, max drive time, terrain preferences, skill level. User context provides date/location for resolving "today/tomorrow".
AssessConditions	Enriched candidates	Day quality assessment	Compare conditions across all mountains to determine if it's a powder day, icy day, or skip day. Provides context for scoring.
ScoreMountain	Each candidate + assessment	Score (0-100) + pros/cons	Score each mountain considering conditions, terrain match, drive time. Uses day assessment to calibrate (e.g., lower bar on icy days).
GenerateRec	Scores + crowd context	Final recommendation	Produce natural language recommendation with top pick, alternatives, and caveats.

Why AssessConditions matters: Without comparing across mountains, the agent can't distinguish "great day at Mountain X" from "Mountain X is the best of bad options". It enables responses like "Skip today, conditions are poor everywhere" or "Any of these would be great - pick based on drive time."

2. ReAct (Flexible Agent)

Autonomous loop - agent decides which tools to call and when to stop (max 8 iterations):

flowchart TB
subgraph Context
Q[Query]
D[Date]
L[Location]
end
Q --> UC[user_context string]
D --> UC
L --> UC
UC --> SR[SkiRecommendation Signature]
subgraph ReAct Loop
SR --> T[Thought]
T --> A[Action]
O[Observation]
O --> T
end
subgraph Tools
A -.-> DB[(search_mountains)]
A -.-> W[get_mountain_conditions]
A -.-> R[get_driving_time]
A -.-> CL[check_crowd_level]
end
DB -.-> O
W -.-> O
R -.-> O
CL -.-> O
T --> |Finish| F[Recommendation]

Key differences from Pipeline:

No ParseSkiQuery - agent interprets query directly in its reasoning
No AssessConditions/ScoreMountain - agent does comparative reasoning implicitly in Thought steps
Flexible order - agent chooses which tools to call and may skip some or call multiple times
Single signature - just SkiRecommendation(query, user_context) → recommendation

Execution Traces

Save detailed execution traces with --save-trace for debugging and comparison:

# Compare both agents on the same query
python -m powder --date 2025-03-29 --save-trace --pipeline "Worth skiing today?"
python -m powder --date 2025-03-29 --save-trace "Worth skiing today?"

Example traces:Pipeline | ReAct

Trace Contents	Pipeline	ReAct
`meta`	query, agent, date, location, model, timestamp	same
`result.parsed`	Structured filters from ParseSkiQuery	-
`result.candidates`	7 mountains with conditions + drive times	-
`result.day_assessment`	Day quality, best available, context	-
`result.scores`	7 scored mountains with pros/cons	-
`result.recommendation`	-	Final text output
`lm_history`	10 LLM calls (Parse + Assess + 7×Score + Generate)	4 LLM calls showing Thought→Action→Observation loop

The ReAct trace shows the agent's reasoning process (skip day detection):

Call 1: "The user is asking if it's worth skiing today... Let me start by searching for nearby mountains"
Call 2: "Nashoba Valley has essentially no snow (0 inches)... Let me check Wachusett Mountain"
Call 3: "Wachusett Mountain also has no snow... Both mountains have essentially no base"
Final: "Not Recommended for Today - insufficient snow coverage"

Setup

make # Create venv and install dependencies
make seed-db # Populate the mountain database
make test# Run tests (fast, no API calls)

Requires uv for package management.

Adding More Mountains and Historic Weather Coverage

Use the seed script to add mountains from the predefined list via Claude slash comand add_mountain.md. Update the hardcoded list in the script if you want to expand to other regions, etc.

# List mountains not yet in the database
python scripts/seed_missing_mountains.py
# Add N mountains using Claude to research and populate data
python scripts/seed_missing_mountains.py --run --num 10
# After adding mountains, re-fetch historic weather data
make fetch-historic

Add API keys to .env:

ANTHROPIC_API_KEY=your_key_here
OPEN_ROUTE_SERVICE_API_KEY=your_key_here # Get free key at https://openrouteservice.org

Usage

# Run with a query (uses ReAct agent by default)
python -m powder "Where should I ski tomorrow? I have an Ikon pass."# Use Pipeline instead of ReAct
python -m powder --pipeline "Best powder within 3 hours?"# Run on historic data (uses fixtures instead of live weather API)
python -m powder --date 2025-02-17 "Best powder day with Ikon pass?"# Different starting location
python -m powder --location nyc "Epic pass, best terrain?"# Interactive mode
python -m powder

Historic Sanity Checks

Test the agent on known good/bad days from the 2024-2025 season:

Powder Days (clear winners):

Date	Scenario	Expected	Command
2025-02-17	Big dump	Sugarloaf (5.7")	`python -m powder --date 2025-02-17 "Best powder with Ikon?"`
2025-03-29	Late season storm	Stowe (5.5")	`python -m powder --date 2025-03-29 "Epic pass powder?"`

Skip Days (bad conditions):

Date	Scenario	Conditions	Command
2025-01-08	Brutal cold	-8°F, 0.4" avg	`python -m powder --date 2025-01-08 "Worth skiing today?"`
2024-12-22	Pre-Xmas ice	-7°F, no fresh	`python -m powder --date 2024-12-22 "Ikon pass today?"`
2025-03-31	Spring slush	54°F avg, 69°F max	`python -m powder --date 2025-03-31 "Should I ski today?"`

Beautiful Days (multiple good options):

Date	Scenario	Conditions	Command
2025-01-29	Warm refresh	1-2" everywhere, 25°F	`python -m powder --date 2025-01-29 "Best skiing today?"`
2025-02-03	Tie-breaker	1" avg, pleasant	`python -m powder --date 2025-02-03 "Where should I go?"`

These dates have known conditions - useful for validating agent behavior on different scenarios.

Evaluation

The evaluation framework measures agent performance with deterministic metrics (no LLM-as-judge).

Current Results (GEPA-Optimized)

Metric	Pipeline	ReAct
Hit@1	93.8%	87.5%
Hit@3	93.8%	87.5%
Constraint Satisfaction	100.0%	n/a*

*ReAct constraint metric is n/a because it returns unstructured text.

Pipeline improved from 41.7% → 93.8% Hit@1 through Pydantic fixes and GEPA optimization.

Metrics

Metric	Description
Hit@1	Top recommendation matches expected mountain
Hit@3	Expected mountain appears in top 3
Constraint Satisfaction	Respects pass type, drive time, terrain requirements
Parse Accuracy	Correctly extracts filters from natural language

Dataset

47 labeled examples across 4 signatures + 16 end-to-end:

ParseSkiQuery - 12 examples (97.4% accuracy)
AssessConditions - 4 examples (100.0%)
ScoreMountain - 8 examples (96.2%)
GenerateRecommendation - 7 examples (100.0% accuracy)
End-to-end - 16 examples with historic weather data

Running Evals

make eval# Run full evaluation suite
make eval-verbose # Show detailed output for failures# Run specific mode
python -m powder.evals.runner --mode react
python -m powder.evals.runner --mode pipeline
python -m powder.evals.runner --mode both

GEPA Optimization

Optimize signatures with GEPA (Reflective Prompt Evolution):

# Optimize individual signatures
python -m powder.evals.optimize --signature score_mountain --max-calls 50
python -m powder.evals.optimize --signature assess_conditions --max-calls 50
# Optimize ReAct with tool descriptions
python -m powder.evals.optimize --signature react --max-calls 30

Optimized prompts are saved to powder/optimized/ and automatically loaded by the pipeline.

Historic Weather Data (Backtesting)

Fetch real weather data from the 2024-2025 ski season for reproducible backtesting:

# Fetch full season (Dec 1 - Apr 15, all mountains × 136 days)
make fetch-historic
# Fetch specific date range
.venv/bin/python -m powder.evals.fetch_historic --start 2025-01-01 --end 2025-01-31
# View summary of fetched data (best powder days, coldest days, etc.)
make backtest-summary

This creates fixtures in powder/evals/fixtures/ that can be used for:

Creating eval examples from real historic conditions
Reproducible backtesting without API calls
Analyzing which days the agent would have recommended correctly

Testing

make test# Fast tests (no API calls)
make test-llm # LLM tests (requires ANTHROPIC_API_KEY)
make test-all # All tests
make test-cov # With coverage report

Project Structure

powder/
├── __main__.py # CLI entry point (python -m powder)
├── pipeline.py # Explicit multi-step pipeline
├── agent.py # ReAct-based agent
├── signatures.py # DSPy signatures
├── tools/
│ ├── database.py # Mountain DB queries
│ ├── weather.py # Open-Meteo client
│ ├── routing.py # Drive time calculations
│ └── crowds.py # Holiday/vacation detection
├── data/
│ ├── mountains.jsonl # Mountain metadata
│ └── mountains.db # SQLite database
├── optimized/ # GEPA-optimized prompts (auto-loaded)
│ ├── parse_query.json
│ ├── score_mountain.json
│ ├── assess_conditions.json
│ └── react_agent.json
└── evals/
├── runner.py # Evaluation runner
├── optimize.py # GEPA optimization script
├── backtest.py # Historic data backtesting
├── find_interesting_days.py
└── fixtures/ # Historic weather data (136 days)

Stack

DSPy - Agent framework with signatures and optimizers
Open-Meteo - Weather/snow data (free, no API key)
OpenRouteService - Drive time calculations (free tier)
SQLite + SQLAlchemy - Mountain database

Future Improvements

Data Sources

Enhancement	Description	Difficulty
Liftie API	Real-time lift status (open/closed/hold)	Easy
Trail grooming reports	Which trails were groomed overnight	Medium
Webcam integration	Visual conditions verification	Medium
Snow report scraping	Official resort-reported conditions	Medium

Snow Quality Inference

Currently we use raw weather data. Could infer surface conditions from patterns:

Weather Pattern	Inferred Quality
Fresh snow + stayed cold	Powder
Rain → overnight freeze	Icy/crusty
Warm afternoon (>35°F)	Slushy/spring
Cold + wind + no new snow	Hardpack/wind-affected
Prolonged cold, no precip	Machine groomed, firm

Smarter Crowd Prediction

Current: Holiday/vacation week detection only.

Future possibilities:

Historical lift line wait times
Parking lot camera analysis
Real-time traffic data on access roads
Weather-based demand modeling (powder days = crowds)

User Personalization

Save user preferences (home location, pass type, skill level)
Learn from feedback on recommendations
"I went to X and it was great/terrible" → adjust scoring

Regional Expansion

Currently: Northeast US (31 mountains)

Could expand to:

Colorado Front Range / Summit County
Utah (Park City, Salt Lake resorts)
Lake Tahoe
Pacific Northwest

Agent Improvements

Feature	Description
Multi-day planning	"Best 3-day ski trip next week"
Group optimization	Balance different skill levels in party
Budget awareness	Factor in lift ticket prices, gas costs
Alternative activities	"If skiing is bad, what else nearby?"

Name		Name	Last commit message	Last commit date
.claude		.claude
anyscale_presentation		anyscale_presentation
docs		docs
evaluation		evaluation
powder		powder
scripts		scripts
tests		tests
traces		traces
.gitignore		.gitignore
CLAUDE.md		CLAUDE.md
Makefile		Makefile
PLAN.md		PLAN.md
README.md		README.md
pytest.ini		pytest.ini
requirements.in		requirements.in
requirements.txt		requirements.txt

Latest commit

History

Folders and files

Repository files navigation

Powder

Features

Query Understanding

Mountains Covered

Data Sources

Context (Provided at Query Time)

Database (Static Mountain Metadata)

Tools (Dynamic API Calls)

Architecture

Tools:

1. Pipeline (Explicit Flow)

2. ReAct (Flexible Agent)

Execution Traces

Setup

Adding More Mountains and Historic Weather Coverage

Usage

Historic Sanity Checks

Evaluation

Current Results (GEPA-Optimized)

Metrics

Dataset

Running Evals

GEPA Optimization

Historic Weather Data (Backtesting)

Testing

Project Structure

Stack

Future Improvements

Data Sources

Snow Quality Inference

Smarter Crowd Prediction

User Personalization

Regional Expansion

Agent Improvements

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages