Available for Contract & Full-time Roles · AI/ML engineering · RAG systems · retrieval architecture · UK-based, remote-friendlyGet in touch →
Compare semantic search and graph traversal on the same query, then see the auto-router explain why it chose each method.
Real enterprise knowledge looks like this: incident reports, supplier updates, customer feedback, staffing memos — scattered across systems, written by different people, never designed to be queried together.
BrewPulse Coffee simulates exactly that. The only way to answer "why did Leeds underperform in Q1?" from these eight documents is to read all of them manually — or use a retrieval system like this one (RAG).
20 ready-made questions · one click to run
201
entities
175
relationships
8
documents
20
sample queries
Built on production-grade AI infrastructure
Real enterprise knowledge looks like this: incident reports, supplier updates, customer feedback, staffing memos — scattered across systems, written by different people, never designed to be queried together.
BrewPulse Coffee simulates exactly that. The only way to answer "why did Leeds underperform in Q1?" from these eight documents is to read all of them manually — or use RAG.
Synthetic corpus, manually curated. Full methodology and engineering decisions on the dataset page.
02_incident_leeds.md
Espresso Machine Failure — Leeds Central
Formal incident log INC-2024-0312. The technical anchor for cross-document reasoning.
03_supplier_northbrew.md
NorthBrew Supplies Oat Milk Disruption
Wakefield depot logistics failure. Documents a prior Sep 2023 incident — critical for recurrence detection.
04_customer_feedback.md
Customer Feedback Summary
47 submissions, 61% negative. "Watery espresso" complaints — never explicitly linked to the valve fault.
08_logistics_mobile.md
Mobile Ordering Rollout Disruption
Orda POS modifier sync bug. Bridges the supplier thread and technology thread across the corpus.
201 entities, 175 relationships, 8 documents — follows the chain.
Same query, two methods, side by side. See which one wins.
The router explains its choice — signals, entity count, confidence.
THE CENTRAL CAUSAL CHAIN — QUESTION 20 DEMO HIGHLIGHT
Wakefield depot systems migration fails
file 03
40% oat milk shortfall at Leeds
files 01, 03
Mobile oat flat white order unfulfillable
file 08
Formal complaint + Google Review posted
file 04
No single document tells this story. Graph traversal reconstructs it; semantic search misses hops 2 and 3.
NorthBrew Supplies experienced delays at the Wakefield depot in early Q1, impacting oat milk deliveries to North England branches. Customer feedback noted supply-related complaints across several locations.
files: 01, 03, 07
↳ fragments retrieved, chain incomplete
The Wakefield depot systems migration (file 03) caused a 40% oat milk shortfall at Leeds. This made the oat flat white unfulfillable via mobile ordering (file 08), generating a formal complaint and a negative Google Review (file 04).
files: 01, 03, 04, 08
↳ full 4-hop chain reconstructed
↳ Graph found file 04 (Google Review complaint) and file 08 (mobile order) via entity chain — semantic missed them because "Google Review" doesn't appear in file 03.
Same query, two methods — graph reconstructs what semantic misses.
ANTHROPIC CERTIFIED
This project built with Claude Code agentic workflow · CLAUDE.md context preservation · custom extractor via Claude tool use