🏎️ F1 Knowledge Agent

Navigate F1 history with AI — grounded answers, zero hallucination, exportable data.

The F1 Knowledge Agent is a RAG (Retrieval-Augmented Generation) agent grounded in Formula 1 Teams & Constructors data. Unlike a live-search agent, it answers questions only from its pre-loaded knowledge base — making every answer reliable, citeable, and auditable.

Every response comes in two layers:

• 📖 Plain English answer — sourced directly from documents, with citation
• 📊 Data Output table — structured data ready to copy into Excel, Python, or Tableau

---

Architecture

User Question → search_documents tool → Relevant passages → Claude → Answer + Data Table

The RAG Equation

Component	Implementation
Brain	Claude Sonnet 4 (Anthropic) — native tool use pattern
Knowledge Base	6 curated F1 documents loaded as Python dictionaries
Retrieval	Keyword search with title weighting (2x) and stop-word filtering
Output	Two-layer: plain English answer + exportable markdown table

How it differs from Compass

	Compass (Live Search)	F1 Knowledge Agent (RAG)
Data source	Live web, Wikipedia, Google News	Pre-loaded static knowledge base
Best for	Current events, real-time data	Historical facts, structured analytics
Reliability	Depends on live sources	Fully auditable — answers trace to docs
Architecture	ReAct loop with 4 external APIs	RAG retrieval with native tool use

Together, these two projects demonstrate both core AI agent architectures used in production systems.

---

Knowledge Base (6 Documents)

#	Document	Coverage
1	Ferrari Scuderia History	Founded 1929, 16 titles, Schumacher era, current status
2	Red Bull Racing History	Founded 2005, Vettel dynasty, Verstappen dominance, 2023 peak
3	Mercedes AMG Petronas History	8 consecutive titles, Hamilton era, post-2021 decline
4	McLaren Racing History	Senna/Prost golden era, decline, 2024 comeback
5	Williams Racing History	Frank Williams dynasty, financial struggles, Dorilton sale
6	Constructors Championship All-Time Data	Full year-by-year table 1958–2024, all team totals

Document 6 is the analytics layer — structured as a clean table so the agent can produce exportable data outputs for further analysis in Python, Excel, or Tableau.

---

What the Agent Can Answer

Easy (direct facts):

• "Which F1 team has the most Constructors Championships?"
• "When was Red Bull Racing founded?"
• "How many titles has Mercedes won?"

Medium (comparative):

• "Compare Red Bull and Mercedes titles won after 2010."
• "Which team won back-to-back Constructors titles most recently?"

Hard (analytical):

• "Which team had the most dominant single season in F1 history?"
• "What pattern do teams show after a dominant era ends?"

Gracefully declined (knowledge gaps):

• "Who will win the 2026 title?" → Agent refuses to speculate
• "What was Ferrari's tire strategy in the 2022 Bahrain GP?" → Agent flags the gap

---

RAG Failure Modes (Task 2)

The notebook deliberately stress-tests three core RAG failure scenarios every production system faces:

Failure Mode	Test	Key Lesson
Outdated data	Inject 2019 standings alongside 2024 docs	Keyword search has no date awareness
Contradicting documents	Two sources give different Ferrari title counts	Agent blends without flagging conflict
Knowledge gap	Ask for race-level data not in any document	Risk of hallucination when retrieval is partial

Each failure includes a written root-cause analysis and proposed fixes — the same evaluation framework used in real enterprise RAG deployments.

---

Quick Start

1. Open F1_Knowledge_Agent.ipynb in Google Colab
2. Get an Anthropic API key at console.anthropic.com
3. Paste your key in Cell 2
4. Runtime → Run all

---

Sample Output

QUESTION: Which F1 team has the most Constructors Championships?
============================================================
  🔍 Searching: 'most constructors championships F1 team'
     Found 3 relevant documents

📖 ANSWER:
Ferrari holds the record for the most Constructors Championships in Formula 1
history with 16 titles, spanning from 1961 to their most recent in 2008.
Source: constructors_championship_alltime.md, ferrari_history.md

📊 DATA OUTPUT:
| Team      | Constructors Titles | First Title | Last Title | Active |
|-----------|--------------------:|-------------|------------|--------|
| Ferrari   | 16                  | 1961        | 2008       | Yes    |
| Williams  | 9                   | 1980        | 1997       | Yes    |
| McLaren   | 9                   | 1974        | 2024       | Yes    |
| Mercedes  | 8                   | 2014        | 2021       | Yes    |
| Red Bull  | 6                   | 2010        | 2023       | Yes    |

---

Tech Stack

• LLM: Claude Sonnet 4 (Anthropic API, native tool use)
• Retrieval: Keyword-based search with title weighting and stop-word filtering
• Domain: F1 Teams and Constructors Championship (1958–2024)
• Environment: Google Colab
• Course: ISOM 260/900 — AI for Business, Suffolk University

---

Related Projects

Project	Description
Compass Agent	Live-search AI research agent using Wikipedia, web fetcher, calculator, and Google News
Sports Analytics Portfolio	Predictive modeling and statistical analysis across multiple sports domains

---

Author

Benedict Daxell Santoso

• Business Analytics & Information Systems, Suffolk University (May 2026)
• LinkedIn: linkedin.com/in/benedictdaxellsantoso
• GitHub: github.com/bdsantoso168

---

🏎️ F1 Knowledge Agent — grounded answers, exportable data, zero hallucination.