Files
sure/app/models/insight/generator.rb
Claude d25bb6e2a3 Address all CodeRabbit PR review comments on insights feed
- Add `Setting.insights_enabled` feature gate at top of GenerateInsightsJob#perform
- Fix race condition in upsert_insight with recursive retry on RecordNotUnique
- Use Provider::Openai.effective_model instead of hardcoded DEFAULT_MODEL
- Rescue Money::Currency::UnknownCurrency in currency_symbol fallback
- Fix budget dedup key to stable "budget_pacing:<date>" (prevents flip-flop between at_risk/on_track)
- Fix cash flow projection off-by-one: iterate (1..PROJECTION_DAYS) not (0..PROJECTION_DAYS)
- Eliminate N+1 in IdleCashGenerator by precomputing active_account_ids in one SQL query
- Rename all_time_high -> thirty_day_high (series is 30-day, not truly all-time)
- Update net_worth_milestone i18n: title_ath -> title_30d_high, "30-day high" language
- Anchor SpendingAnomalyGenerator baseline to current_period.start_date (not calendar months)
- Fix spending_anomaly dedup key to use current_period.start_date instead of Date.current
- Add .order(last_occurrence_date: :asc) to SubscriptionAuditGenerator for deterministic results
- Stub Setting.insights_enabled in job tests; add generated_at assertion to upsert test

https://claude.ai/code/session_014vY9xohpm3abSAxVxRF27a
2026-04-12 14:17:38 +00:00

67 lines
1.8 KiB
Ruby

# Base class for all insight generators.
#
# Subclasses must implement #generate, returning an Array<GeneratedInsight>.
# Financial reasoning is done in pure Ruby using existing analytics infrastructure.
# The LLM is invoked only to write the human-readable body text.
class Insight::Generator
GeneratedInsight = Data.define(
:insight_type,
:priority,
:title,
:body,
:metadata,
:currency,
:period_start,
:period_end,
:dedup_key
)
attr_reader :family
def initialize(family)
@family = family
end
def generate
raise NotImplementedError, "#{self.class.name} must implement #generate"
end
private
def llm
@llm ||= Provider::Registry.get_provider(:openai)
end
# Generates a 1-2 sentence natural-language explanation using the LLM.
# Falls back to a bare template string if no LLM is configured.
def generate_body(prompt)
return prompt unless llm
response = llm.chat_response(
prompt,
model: Provider::Openai.effective_model,
instructions: system_instructions
)
response.messages.first&.output_text&.strip.presence || prompt
rescue => e
Rails.logger.warn("[Insight::Generator] LLM body generation failed: #{e.message}")
prompt
end
def system_instructions
sym = currency_symbol
<<~PROMPT
You are a concise financial insights writer for a personal finance app.
Write exactly 1-2 sentences in plain, conversational English.
Be specific with numbers. Use #{sym} for currency amounts.
Do not use jargon, emoji, or give investment advice.
PROMPT
end
def currency_symbol
Money::Currency.new(family.currency).symbol
rescue Money::Currency::UnknownCurrency
family.currency
end
end