01
Ingest
Telemetry flows in.
FastF1 session data — laps, sectors, tire compounds, gaps, weather — assembled into a structured race-state snapshot. Replay any 2024 GP at 1×, 4×, or 16×.
Race strategy that shows its work.
APEX uses IBM Granite 4.0 to analyze telemetry and generate pit strategy recommendations — with the full reasoning chain visible. Every call is transparent. No black box.
★ Explainability-first · Every AI decision shows 4 data pointsLAP
32/78
POSITION
P1
↑ Leader
GAP
+1.812
↓ closing
TIRE HARD
71%
↓ −0.04s/lap deg
Gap to leader last 32 laps
LEC
PIA
APEX · Race Engineer
LIVE · Granite
Current call — Lap 32
BOX BOX
Undercut window closes in 4 laps.
Tire deg
HARD age 31L · +0.04s/lap, cliff at L38
Undercut
PIA gap 1.8s — pit-loss 22.4s nets +1.1s
Rejoin
Clean air confirmed. No TSU traffic
Weather
Dry through L48. No forcing function
Confidence
84%
Why undercut now?
Built with
IBM Granite 4.0 H-Small
FastF1
React 18
▲ Vercel
Apache 2.0
The AI Layer
IBM Granite 4.0 is the intelligence at the core.
APEX doesn't bolt AI on as a feature. IBM Granite 4.0 H-Small is the engine that reads the race state, runs the strategy math, and explains its decisions in language engineers actually use.
- Structured JSON output — every call returns severity, confidence, call, reasoning chain, and risk in a validated schema.
- Race-engineer persona — the system prompt defines a senior strategist who speaks F1 vocabulary: undercut, overcut, cliff, pit-loss, delta.
- Telemetry snapshot context — Granite receives live race state: lap, gaps, tire compounds, deg rates, weather, and the pit window before responding.
- Fallback-safe — if the live call fails, canned expert cards maintain the demo without a broken experience.
ibm-granite/granite-4.0-h-small
CONNECTED
1
Telemetry snapshot
Lap, gaps, tire age, deg rate, weather, pit window
2
Race-engineer prompt
System context + structured JSON schema
3
Granite inference
granite-4.0-h-small · streaming disabled
4
Strategy card + reasoning
Severity · call · 4 evidence points · risk · confidence
5
Engineer follow-up
"Why now?" → Granite answers the reasoning question
By the numbers
F1 is decided in data.
APEX reads it for you.
0
SENSORS · PER CAR
Generating a continuous stream through every corner at racing speed.
0M
POINTS · PER SECOND
Telemetry data points streamed from each car to the pit wall every second.
0+
DECISIONS · PER RACE
Strategy calls made across a race weekend — most without clear explanations to anyone outside the pit wall.
0
UNEXPLAINED CALLS
Black-box recommendations in APEX. Every call ships with the 4-point reasoning chain.
How it works
Telemetry in. Strategy out.
Reasoning in between.
Three steps · no black box
02
Reason
Granite reasons.
IBM Granite 4.0 H-Small evaluates the snapshot — undercut deltas, tire-cliff projections, threat windows — against a race-engineer system prompt. Returns structured JSON with severity, call, confidence, and a 4-point reasoning chain.
03
Explain
Every call explains itself.
A pit-wall recommendation in plain English — with the four data points Granite weighed visible on screen. Ask "why now?" and Granite answers in the same race-engineer voice, showing its work in the way a senior strategist would brief a junior engineer.
The differentiator
Most AI predicts. APEX explains.
The IBM SkillsBuild challenge asks for AI that "builds trust through explainability." That's not a feature — it's the entire point. Every strategy card in APEX ships with the evidence that produced it.
Explainability · not a feature · the whole product
Transparency
The call and the why. Together.
APEX never shows "Pit on lap 36" without showing why: tire age, undercut delta, threat assessment, weather. Judges see 4 evidence points per decision.
GRANITE · ADVISORY CARD
CallBOX BOX
Tire deg+0.04s/lap — HARD cliff at L38
Undercut+1.1s vs PIA on out-lap
RiskCover window closes L38
Interrogable
Ask "why." Get an answer.
The engineer panel accepts free-text questions. "Why undercut now?" triggers a Granite call that references the exact telemetry values in its response — not a canned string.
ENGINEER FOLLOW-UP
QWhy undercut now and not L34?
GranitePIA's HARD age matches ours — no defensive cushion. Waiting 2 laps means TSU traffic at T6 on the out-lap, losing the position advantage.
Calibrated
Confidence that means something.
Every card ships with a 0–100% confidence score derived from the strength of each evidence signal. A VSC call at 96% confidence looks different from a weather call at 71%.
CONFIDENCE CALIBRATION
VSC DEPLOYED96% — Act immediately
UNDERCUT84% — Window open
RAIN RISK71% — Monitor
Features
A pit wall, distilled.
Each surface answers a real strategy question. Nothing exists for decoration.
06 surfaces
Explainable AI strategy
Pit calls with reasoning. Not just predictions.
Every strategy card shows severity, confidence, the call, and the 4 data points IBM Granite weighed — making AI decisions trustworthy and legible.
Tire intelligence
Degradation curves, projected.
Lap-by-lap deg slope with cliff-edge projection per compound. Granite references the cliff index in its reasoning chain.
Undercut math
Live delta projection.
Real-time gap math vs every car in strategic range, fed directly into Granite's context.
Weather
Adaptive strategy.
Rain probability, track temp, wind — folded into every Granite call as structured context.
Scenario mode
DRY · UNDERCUT · RAIN · VSC.
Switch scenarios and watch Granite update its recommendation and reasoning in real time.
Open source
Apache 2.0.
The system prompt is public. The model is open. The reasoning is visible. No vendor lock.
Inside the dashboard
Three views.
One race, decoded.
strategy · compare · telemetry
Strategy
Read the race the way engineers do.
A unified canvas of lap times, gap deltas, tire deg, and Granite's verdict — all updating together as you replay the session.
- Critical alerts surfaced as a single, ownable red moment
- Threats from cars in strategic range, named and ranked
- Playback at 1×, 4×, 16× — any 2024 GP on demand
Race engineer · feed
LAP 32 · 14:53:02
Critical
Pit window for LEC closes in 4 laps
Tire deg +0.04s/lap. PIA same tire age. Undercut gain: +1.1s. Clean rejoin confirmed.
LAP 25 · 14:42:11
Advisory
Hold position. Pit window opens L34.
HARD deg within window. Track position value high — Monaco OT EV ≈ 0.06.
Comparison
Where time was won.
And where it slipped.
Two drivers, three sectors, one delta — broken down by Granite into the specific corner or throttle application where the lap was decided.
- AI-narrated comparison summaries per sector
- 24 micro-sector bar chart — see exactly where ms were lost
- Side-by-side stint timeline with pit-stop markers
Sector delta · LEC vs PIA · Monaco
LECLERC
Ferrari · #16
VS
PIASTRI
McLaren · #81
TOTAL
+0.424
S1
−0.034
S2
+0.280
S3
+0.178
IBM SkillsBuild · AI Beyond the Finish Line
Built for the challenge. Genuinely.
The challenge asks for AI that drives smarter decisions, improves understanding, and builds trust through explainability. APEX was designed from the ground up to address all three — not as a checklist, but as the core product thesis.
AI Strategy & Decision Support
Granite generates pit strategy calls with severity, call, confidence, and 4-point reasoning per recommendation.
AI Analytics
Tire deg curves, sector delta charts, gap-to-leader history, and weather forecast — all feeding Granite's context window.
AI Copilot
Free-text engineer input triggers live Granite calls. "Should we pit now?" returns a structured strategy card in real time.
Explainability
Every recommendation is accompanied by the 4 telemetry signals Granite evaluated — satisfying the challenge's core trust requirement.
Ready to read the race?
APEX is open-source, runs on IBM Granite 4.0, and shows every decision's reasoning. Built for the IBM SkillsBuild May Innovation Challenge.