PokerScout

Project Overview

PokerScout is an AI analyst that ingests a Hendon Mob player profile URL and outputs a standardized scouting dossier. It parses the page, extracts key stats and rankings, and models stake-aware performance (last 12 months & lifetime). Output is delivered as a strict JSON schema and a clean HTML report.

Genesis

Born from the need to look beyond “Total Live Earnings,” PokerScout converts a single Hendon Mob URL into an analyst-grade profile that blends stake-aware performance, competitive tiering, rankings, and travel cadence. The result is a production-ready Agent.AI workflow with a strict JSON contract and polished HTML report that standardizes scouting and reduces manual research into a fast, explainable output.

Scope

• Agent workflow design (Agent.AI): Orchestrated the end-to-end action flow (input URL → fetch → parse → analyze → render).
• Prompt engineering: Deterministic LLM prompt with a typed JSON contract, null-safe fields, and rule-based heuristics.
• Information extraction: Patterns to pull player name, totals, min/max cash with years, GPI ranking, Popularity ranking, series counts, and location features from noisy HTML.
• Domain logic & modeling:
  • Stake tier classification from event titles (handling “buy-in + fee” formats & currency hints).
  • Final-table rules (9-max default, 6-/8-max detection).
  • Competitive tiering with High Roller override by average buy-in and high-roller share.
• Data normalization: Standardized stake buckets (Low/Mid/High/High Roller), lifetime vs. 12-month views, percent-of-total calculations.
• Presentation & UX: Consistent HTML sections and reversed stake order (High Roller → High → Mid → Low) for performance blocks; inline definitions and graceful fallbacks (“Unavailable”).
• Quality & robustness: Null-handling, defensive parsing, consistent formatting to prevent broken placeholders.

Skills Demonstrated

• AI/LLM: Prompt engineering, schema design, deterministic extraction, instruction following.
• Data parsing: Regex/pattern matching, HTML text normalization, currency/buy-in interpretation.
• Analytics & modeling: Heuristic modeling for stake segmentation, temporal rollups.
• Product thinking: Clear problem framing (beyond “Total Live Earnings”), explainable outputs, resilience to missing data.
• Frontend output: Semantic HTML templating, readability, accessibility-friendly layout choices.
• Ops & QA: Error handling, guardrails, and reproducibility across profiles.

Deliverables

• Structured JSON schema (playerInsights) with clearly defined fields and enums.
• HTML report template matching the schema:
  • Results Summary (incl. GPI & Popularity)
  • Competitive Profile (Tier, Volume Style + concise descriptions)
  • Activity & Recency; Volume & Consistency
  • Financial Stats (min/max cash + years)
  • Performance (Last 12 Months) & Performance (Lifetime) by stake with cashes / % / wins / FTs (ordered HR → High → Mid → Low)
  • Tournament Series Results (WSOP/WPT/EPT/Regional)
  • Travel & Tour Activity (cities, avg days, clusters)