Skip to main content

Data-Driven AI Chat

Version: 3.2.4 (January 2026) Category: AI & Features

The Vartovii AI Chat has been enhanced to provide data-driven responses that cite actual numbers and facts from our database, rather than speaking hypothetically.

The Problem​

Previously, when asked about tokenomics or investor concentration, the AI would respond with phrases like:

  • ❌ "If there are large token unlocks..."
  • ❌ "If a significant portion is controlled by large holders..."
  • ❌ "Якщо передбачається значне розблокування..."

This was unhelpful because users wanted specific data, not hypothetical scenarios.

The Solution​

1. Enhanced Tool: get_crypto_trust_score​

The AI's function calling tool now fetches complete project data from the database:

Data CategoryFields Returned
Token Unlockstoken_unlocks array, unlock_progress_pct
Investorslead_investors, investors, backer_tier_score
Fundingtotal_raised, funding_rounds_count
Treasurytreasury_usd, runway_months
TVL & DeFitvl, chain_tvls, mcap_to_tvl_ratio
Score Breakdownscore_breakdown (why each pillar scored X)
Securitycertik_score, tweetscout_score

2. System Prompt Instructions​

The AI follows strict rules:

## CRITICAL: DATA-DRIVEN RESPONSES
When answering about crypto projects:
1. NEVER say "if" ("якщо", "если") - state actual facts from the data
2. CITE SPECIFIC NUMBERS: Instead of "if there are large unlocks", say 
   "according to our data, 45.7% of tokens are still locked, with the 
   next unlock of 15M tokens scheduled for March 2026"
3. NAME INVESTORS: Don't say "if large holders control the supply", 
   instead say "Lead investors include a16z and Paradigm, who hold X% of supply"
4. USE score_breakdown: Explain WHY the Treasury score is 30 - 
   "Treasury Score is low (30/100) because TVL is only $X million 
   compared to $Y billion market cap"
5. QUOTE SOURCES: "За даними DefiLlama...", "Дані Dropstab показують..."
6. If data is NULL or unavailable, say "Дані про це недоступні в нашій базі" 
   instead of speculating

Example Responses​

Before (Hypothetical) ❌​

"Якщо значна частина FXS зосереджена в руках кількох великих власників, це може викликати занепокоєння..."

After (Data-Driven) ✅​

"За нашими даними, 55.3% токенів FXS вже розблоковано. Наступний unlock запланований на березень 2026 (15M FXS, категорія Team). Lead investors включають Dragonfly Capital та Mechanism Capital. Treasury Score низький (30/100) через високу експозицію до власного токену FXS у казначействі протоколу."

Token Unlocks Interpretation​

The AI understands the vesting data structure:

FieldMeaning
token_unlocksArray of unlock events (category, raised_m, unlock_pct, status, vesting_period)
unlock_progress_pctWhat % of total supply is already unlocked
High unlock_progress (above 80%)Most tokens circulating = GOOD for tokenomics
Low unlock_progress (below 40%)Many tokens still locked = potential sell pressure

Technical Implementation​

File: backend/ai/services/chat_service.py

Key Functions:

  • _tool_get_crypto_trust_score() - Returns full project data
  • _chat_with_tools() - Enhanced system prompt with data-driven rules

When Data is Unavailable​

If a field is NULL in the database, the AI will honestly state:

"Дані про графік вестингу недоступні в нашій базі. Для отримання цієї інформації рекомендую перевірити Dropstab або офіційну документацію проекту."

This is preferable to speculation or hypothetical statements.