giorgio/read.markets
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

i18n: prepend a strong language directive for portfolio + chat

Browse source 
Reports that portfolio AI analysis was coming back in English even
for IT-toggled users. Traced the chain (DB user.lang IS set to it,
router passes it into the payload, parse_request reads it, build_prompt
appends respond_in_clause), so the wiring is correct end-to-end. The
model was simply ignoring the single-sentence tail nudge: when the
system prompt is hundreds of lines of English and the user message
adds more English context, "Respond in Italian." at the end is easy
to drop on the floor.

Add a new services/i18n.language_directive_lead() that returns a
strong, explicit top-of-prompt block — "# LANGUAGE — write everything
in <X>" plus the verbatim-tickers-and-numbers carve-out — meant to
be PREPENDED so the model anchors on the target language before it
reads the bulk of the instructions. Combined with the existing tail
clause it's belt-and-suspenders: top + bottom of the prompt both
say "in this language".

Applied to portfolio_analysis.build_prompt() and chat.py — the two
surfaces that generate user-facing prose in real time (the strategic
log + indicator summaries get post-hoc translation via translate(),
so the directive isn't needed there).

Empty-string return for en / unknown lang means callers can wire
it in unconditionally; no extra plumbing in i18n callsites.

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
This commit is contained in:
Giorgio Gilestro 2026-05-29 15:21:00 +02:00
parent 736d161990
commit 13dd3a8330
3 changed files with 52 additions and 9 deletions
Download patch file Download diff file Expand all files Collapse all files

28
app/services/i18n.py
View file

@ -46,3 +46,31 @@ def respond_in_clause(lang: str | None) -> str:
if not lang or lang == "en" or lang not in LANGUAGES:
return ""
return f"\n\nRespond in {LANGUAGES[lang]}."
def language_directive_lead(lang: str | None) -> str:
"""Strong, top-of-prompt language directive for callers that
generate user-facing prose in real time (portfolio analysis,
chat) and need the output to actually land in the user's
preferred language. A single tail clause like
``respond_in_clause`` is easy for the model to ignore when the
rest of the prompt + user message are entirely in English; this
leads with an explicit "all output in X" block, kept verbatim
rules for symbols/numbers, and is intended to be prepended to
the system prompt so the model anchors on the target language
before reading the rest. Combined with respond_in_clause at the
tail it gives a belt-and-suspenders defence.
Empty string for English or unknown codes so callers can paste
it in unconditionally.
"""
if not lang or lang == "en" or lang not in LANGUAGES:
return ""
language = LANGUAGES[lang]
return (
f"# LANGUAGE — write everything in {language}\n"
f"All output — section headers, prose, lists, and any inline "
f"labels — must be written in {language}. Do NOT mix English in. "
f"Ticker symbols (AAPL, MSFT, VOD.L), ISO currency codes "
f"(USD, EUR, GBP), and numeric values stay unchanged.\n\n"
)