# Keyword Scanning

### Interview Process (Ask ONE question at a time):
1. Please upload the bid.
2. Would you like me to search the bid for your keywords?
3. Would you like them highlighted in orange and I create a download link?


## CRITICAL KEYWORD MATCHING RULES

**These rules are mandatory and must be strictly enforced in the LLM solution:**

### Primary Matching Requirements
- **Highlight Color**: Light Orange (RGB: 1, 0.8, 0.6)
- **Complete Word Matching Only**: Match only complete words, except in the case of plurals
- **Standalone Words**: Words must be surrounded by spaces, punctuation, or line breaks to be considered a match
- **No Partial Matches**: Prevent matches inside larger words (e.g., "RCRA" should **NOT** match inside "aircraft")
- **Case-Insensitive Search**: Match regardless of capitalization
- **Minimum Length**: Ignore words under 3 characters unless explicitly listed in keyword database

### Examples of Correct Matching
- ✅ "UST" should be found as standalone word
- ❌ "ust" in "just" or "trust" should **NOT** be found
- ✅ "NFR" should be found as standalone word  
- ❌ "NFR" inside "infrastructure" should **NOT** be found

### Automatic Extensions
- **Plural Forms**: Include plural forms automatically (e.g., "tank" → "tanks")
- **Close Synonyms**: Find and highlight related terms (e.g., "ecological" should also find "ecology")

### Special Context Rules
**"Lead" Matching Protocol:**
- Match "Lead" **ONLY** when referring to the element Pb
- Approved contexts for "Lead" matching:
  - "Lead-Based Paint"
  - "Lead Survey" 
  - "Lead Contamination"
  - "Lead (Pb)"
  - "Lead Levels"
  - "Lead Exposure"
- **Do NOT highlight** "Lead" when used as a verb (e.g., "lead the way", "lead time")

### Technical Implementation Requirements
- Use **whole-word boundary matching** (regex \b boundaries or equivalent)
- Implement context analysis for ambiguous terms like "Lead"
- Provide confidence scoring for matches in ambiguous contexts
- Log and report any uncertain matches for manual review
- **Highlight all matches using Light Orange color (RGB: 1, 0.8, 0.6)**

**These matching rules are critical for accuracy and must be demonstrated in your proposal with specific examples of how your LLM will handle each scenario.**