LLM Match Studio

1 · Upload Excel file

Required columns: right_name and left_name

2 · Select model

Amazon Nova Micro eu-west-1

Fastest and most cost-efficient model. Ideal for high-volume batch runs.

Input $0.035 / 1M tokens Output $0.140 / 1M tokens

Amazon Nova Lite eu-west-1

Low cost multimodal model. Good balance of speed and accuracy.

Input $0.060 / 1M tokens Output $0.240 / 1M tokens

Amazon Nova Pro eu-west-1

Most capable Amazon model. Best accuracy for complex multilingual names.

Input $0.800 / 1M tokens Output $3.200 / 1M tokens

Meta LLaMA 3.3 70B us-east-1

Large multilingual model. Strong on transliterations and cultural name variants.

Input $0.720 / 1M tokens Output $0.720 / 1M tokens

Meta LLaMA 3.1 8B us-east-1

Compact and very fast open-source model. Very cheap for quick batch testing.

Input $0.220 / 1M tokens Output $0.220 / 1M tokens

3 · Prompt template Edit ▼

Use {screened_name} (right name) and {matched_name} (left name) as placeholders. The model must return one word: exact · high · medium · low · none

You are matching two names. Output ONLY one label: exact, high, medium, low, none.

Names:
1) {screened_name}
2) {matched_name}

Rules (apply in order):

A) EXACT:
Return exact if names are identical after ignoring:
- case, punctuation, accents
- organization legal suffixes (inc, ltd, gmbh, llc, s.a., ag, bv, oy, kg, srl, spa)

B) HIGH:
Return high if core name tokens strongly match with only small changes:
- same tokens different order
- 1-2 small typos / missing middle token
- known abbreviation expansion (intl/international, co/company)
For people: surname must match OR be a clear transliteration.
For orgs: main root tokens must match (not only suffixes).

C) HARD NON-MATCH OVERRIDES (force low or none):
If any apply, do NOT return high/exact:
- only one short token overlaps (e.g., "Ali")
- different surnames (for people) with no transliteration evidence
- org vs person mismatch
- names share only initials or vague phonetic similarity

D) MEDIUM vs LOW vs NONE:
- medium: plausible match but missing/extra tokens or moderate spelling differences; needs review
- low: superficial similarity; likely different
- none: clearly different; no meaningful token/root overlap

Return ONLY: exact | high | medium | low | none

Analyzing names…

LLM Match Studio