FERRARI DRS
KM/H
287
GEAR
7
LAP
32/78
RPM 11,842

PITWALL

F1 Race Strategy Copilot

SKIP โ€บ
๐Ÿ”Š Click anywhere for engine sound
IBM SkillsBuild AI Builders Challenge โ€” May 2026

PITWALL

F1 Race Strategy Copilot

AI-powered pit stop strategy using real F1 telemetry, IBM Granite reasoning, Docling regulation parsing, and a Q-Learning agent trained on 23,400 real race decisions.

23,400
RL Training Decisions
5
Real F1 Races
3
AI Systems
2022โ€“24
Seasons Covered
IBM
Granite + Docling

How PitWall Works

Three AI systems working in concert โ€” real telemetry data, regulation-aware language reasoning, and reinforcement learning trained on actual race outcomes.

01 โ€” DATA
๐Ÿ“ก
FastF1 Telemetry
Official F1 timing API. Lap times, tyre compounds, pit stop laps from 2022โ€“2024 seasons.
02 โ€” RULES
๐Ÿ“‹
Docling + FIA Regs
IBM Docling parses FIA Sporting Regulations PDF, extracting compound rules and pit window constraints.
03 โ€” REASON
๐Ÿง 
IBM Granite 4.1 8B
Granite receives telemetry + regulation context, generates explainable strategy analysis.
04 โ€” LEARN
๐Ÿค–
Q-Learning Agent
RL agent trained on 23,400 real decisions. Outputs Stay Out / Pit Now / Pit in 2 with Q-values.

Race Strategy Analysis

Select any season (2022โ€“2024), Grand Prix, and driver. FastF1 loads real telemetry. IBM Granite โ€” with Docling regulation context โ€” analyzes the complete race strategy.

  • Real FastF1 telemetry โ€” not simulated data
  • FIA regulation context injected via Docling
  • 3-paragraph strategic breakdown from Granite
  • Identifies optimal vs suboptimal decisions
  • Suggests alternative strategies
Sample Granite Output โ€” Monaco 2024, LEC
Driver: LEC | Race: Monaco GP 2024 Laps: 78 | Best: 75.162s Compounds: MEDIUM โ†’ HARD โ”โ” Strategy Assessment โ”โ” Safety Car pit on lap 1 was textbook โ€” free stop, zero track position loss. โ”โ” Regulation Check โ”โ” Two-compound rule: satisfied โœ“ Pit window timing: optimal โœ“

Live Pit Window Advisor

Adjust sliders for current lap, tyre age, and lap time delta. IBM Granite gives a direct pit/stay-out recommendation simulating real pit wall decision support.

  • Real-time Granite recommendation
  • Tyre compound context aware
  • Lap delta degradation analysis
  • 2โ€“3 sentence direct reasoning
  • Simulates actual pit wall workflow
Sample Output โ€” Lap 25, MEDIUM, +0.5s
Input State: Lap: 25 | Compound: MEDIUM Tyre Age: 20 laps | ฮ”: +0.5s Granite Recommendation: STAY OUT โ€” delta of +0.5s is within acceptable window. With 20 laps remaining, pitting now risks undercut exposure.

RL Pit Optimizer

The Q-learning agent โ€” trained on 23,400 real F1 lap decisions โ€” outputs a pit recommendation with full Q-value transparency. Not a black box.

  • Trained on 5 real F1 races (2023โ€“2024)
  • 23,400 lap state โ†’ decision pairs
  • Full Q-value breakdown shown
  • Confidence score from Q-value margin
  • Explainable โ€” see exactly why it decides
Sample RL Output โ€” MEDIUM, 25 laps old
State: Compound: MEDIUM | Age: 25 Laps Remaining: 30 | ฮ”: +0.8s Q-Values: Stay Out โ†’ 7.600 Pit Now โ†’ 14.978 โ—€ MAX Pit in 2 โ†’ 7.506 Decision: PIT NOW | Conf: 65.2%

Built with IBM AI

PitWall uses two core IBM technologies as required by the challenge โ€” Granite for reasoning and Docling for document intelligence.

Granite 4.1 8B

IBM's flagship open-source language model. Core reasoning engine for strategy analysis and pit window recommendations. Receives telemetry + regulation context, outputs explainable strategic analysis.

Docling

IBM's document intelligence library. Parses the FIA Sporting Regulations PDF, extracts tyre compound rules and pit stop regulations, injects this context into every Granite prompt.

FastF1 + RL Agent

Official F1 telemetry API for real race data (2022โ€“2024). Combined with a custom Q-Learning agent trained on 23,400 real lap decisions from 5 F1 races.

Try PitWall Live

The full application running live on HuggingFace Spaces. Select Monaco 2024 + LEC and click Analyze Strategy to see IBM Granite in action.

huggingface.co/spaces/ashish-doing/pitwall-f1-copilot

Live Race Simulation

Customize your strategy โ€” compound, pit lap, degradation rate. Watch the RL agent make lap-by-lap pit decisions. Pause anytime to analyze Q-values.

Monaco Grand Prix โ€” Leclerc
โ˜€๏ธ Dry ยท Track: 42ยฐC ยท Air: 26ยฐC ยท Wind: 8km/h
โ— READY
GAP: โ€”
Start Compound
Pit Lap: 40
Deg Rate: 3x
2nd Compound
MONACO CIRCUIT โ€” LAP TRACKER
โ–  S1 โ–  S2 โ–  S3
โ€”
PIT LANE STE DEVOTE BEAU RIVAGE CASINO LOEWS TABAC S/F 16 LAP 1/78 LAP TIME 1:15.2 POS P3
Tyre Wear
100%
1
Lap
MEDIUM
Compound
0
Tyre Age
+0.00s
Delta
P3
Position
RL Agent Decision
READY
Awaiting simulation start...
Confidence
โ€”
Q-Values
โ€”