About the Customer
Brand Institute is one of the world's leading pharmaceutical brand naming agencies, responsible for developing and vetting brand names for drugs, biologics, and medical devices. Their linguistic review process requires evaluating candidate names for unintended negative connotations, taboo associations, and phonetic similarities to harmful words — across dozens of languages and global markets.
With hundreds of candidate names moving through their pipeline annually, the manual effort required for initial linguistic pre-screening was a significant bottleneck. Brand Institute engaged Atayo to explore whether AI could accelerate and standardize this process.
Customer Challenge
Screening Pharmaceutical Brand Names Across Languages at Scale
Pharmaceutical brand naming carries serious regulatory and reputational risk. A name that sounds innocuous in English may carry a vulgar, offensive, or death-related meaning in another language — or when spoken aloud in a target market. Brand Institute's linguists were performing this analysis manually, reviewing names against internal knowledge and reference materials across dozens of languages.
The core challenges were:
- Scale and speed: Manual review couldn't keep pace with the volume of candidate names entering the pipeline.
- Consistency: Different reviewers applied different thresholds, making it difficult to standardize risk verdicts.
- Coverage: Ensuring thorough coverage across phonetic variants, letter-string substrings, and regional dialects required deep expertise that was difficult to scale.
- Explainability: Clients needed clear, documented rationale for any flagged name — not just a pass/fail result.
Brand Institute needed a tool that could perform fast, consistent, and explainable pre-screening — freeing their expert linguists to focus on borderline cases and final review rather than initial triage.
Partner Solution
A Multi-Tiered AI Linguistic Analysis Engine on AWS
Atayo designed and delivered a Proof of Concept (POC) for a Linguistic Connotation Screening Tool on AWS, built around a multi-tiered analysis engine that combines deterministic lexicon matching with Generative AI contextual analysis.
Tier 1 — Lexicon Matching
The foundation of the engine is a curated multilingual lexicon database stored in Amazon DynamoDB. Each entry captures a term, its severity rating (A = critical/taboo, B = strong negative, C = mild/awkward), category (vulgar, insult, death, failure, etc.), and language-specific notes. When a candidate name is submitted, the system:
- Normalizes the input (Unicode NFKC, lowercase, diacritic removal)
- Generates n-gram variants (prefixes, infixes, suffixes) to catch embedded substrings
- Runs direct, fuzzy (Levenshtein distance ≤2), and phonetic matching against the lexicon
- Flags any matches with their severity, match type, and rationale
The lexicon database also serves as a cost-control mechanism — terms identified through prior AI analysis are stored and reused, reducing the need to re-process known matches through the LLM on subsequent runs.
Tier 2 — Generative AI Contextual Analysis
For names that pass or partially pass Tier 1, the system invokes Claude via Amazon Bedrock to perform deeper contextual and cultural analysis. Claude assesses:
- Cultural associations and regional connotations not captured in the lexicon
- Phonetic similarity to problematic words when spoken aloud in the target language
- Auto-correct and keyboard-neighbor swap scenarios (e.g., what a search engine might suggest)
- Positive or neutral associations worth noting alongside any risks
Risk Scoring and Verdict Assignment
Each name receives a composite risk score based on severity of matches, match type (exact, fuzzy, phonetic, AI-identified), and language importance weighting. Scores map to one of three verdicts:
- Clear — no significant issues identified
- Investigate — borderline matches requiring human linguist review
- Block — high-severity matches recommending the name be retired from consideration
User Interface and Reporting
A lightweight web UI allows analysts to submit individual names or small batches, select target languages, and view results in real time. Each result includes a per-language breakdown table with triggered terms, match types, severity ratings, and plain-English rationale. Results are exportable as CSV, JSON, and PDF for client delivery and internal documentation.
The architecture was deployed on AWS using Amazon API Gateway and AWS Lambda for the serverless processing layer, Amazon S3 for report storage, Amazon DynamoDB for the lexicon and results database, and Amazon Bedrock for Claude model access — all within a single POC environment with IAM-based access controls.
About the Partner
Atayo Group is an AWS Advanced Consulting Partner specializing in cloud migration, managed infrastructure services, cloud security, and AI and data intelligence. Headquartered in Tampa, FL, Atayo delivers end-to-end cloud solutions for healthcare, financial services, professional services, and enterprise customers across North America. Atayo holds AWS competencies in Migration and Healthcare, and is recognized as an AWS Well-Architected Partner.