Reconsidered pure Go (would mean reimplementing and re-verifying the
already-tested AI pipeline) and pure Python (leaves perceived
performance/concurrency headroom on the table for a preemptive
scale bet). Landed on: Go handles auth/messaging/DB, Python keeps
owning generate_draft/escalation_filter/retrieve_style behind an
internal API. backend/ (Python) is now labeled a verified prototype
for the Go rewrite to match, not the final service.
Native wasn't actually required for the v2 OS-layer notification
listener -- Flutter reaches it via platform channels like any other
native Android API, same pattern many production apps already use.
Flutter's faster iteration on the chat UI (already validated in the
click prototype) matters more for v1 than starting native. Updates
tech-design.md §8, roadmap.md's checklist, and AGENTS.md accordingly.
tech-design.md §8 settles the stack (Android/Kotlin, Python/FastAPI,
PostgreSQL, WebSocket relay, Room+SQLCipher) so it stops blocking
item 1 of the build order. roadmap.md's Phase 1 breakdown is now
checkboxes instead of prose, and AGENTS.md adds the rule to check/
update that checklist before and after any Phase 1 app-build task,
rather than tracking progress ad hoc.
blind_eval.py runs PoC #1's blind-eval methodology over N held-out
corpus dialogues automatically (generate_draft.py refactored to
expose draft_reply() so both share the same drafting logic).
retrieve_style.py implements the keyword/recency search from
tech-design.md §2-1 and wires into generate_draft.py as --history,
replacing hand-curated --style files.
Bash access was intermittently restricted for part of this session
(auto-mode safety classifier), so these were initially written and
committed-pending without live execution. Now verified for real:
generate_draft.py's existing behavior still holds after the
draft_reply() refactor, blind_eval.py runs cleanly against val.jsonl,
and retrieve_style.py's original weighted-sum scoring had a real bug
-- recency drowned out keyword overlap for short Korean messages
(particle attachment means "핀란드" and "핀란드는" don't share a
token), so it was effectively returning the most recent messages
regardless of topic. Fixed by ranking on (overlap, recency) instead
of a weighted sum, confirmed the Finland-related exemplar now ranks
first for a matching query.
escalation_filter.py implements tech-design.md §3's first step as an
actual hard gate, not just a system-prompt instruction: money,
appointment-confirmation, and emotional content stop generate_draft.py
before it ever calls Gemini. Self-test 10/10; measured a 0.93% trigger
rate against 82,305 real corpus utterances (mostly factual price
mentions, not personal money requests -- noted as an upper bound, not
a real-usage estimate).
v1 doesn't train a custom model: it retrieves the closest-matching
past messages from the person's own history and feeds them as
few-shot exemplars to the same prompt contract generate_draft.py
already implements, via the hosted Gemini call. Narrows the AI-Hub
base corpus's role to evaluation and future on-device distillation,
since a hosted LLM already covers general Korean fluency.
Resolves the open Q1-Q7 questions as tentative decisions (self-app beta
first, target consumers, MVP scenario = read-receipt relief + group-chat
catch-up, autonomy capped at L0-L2) and builds the standard deliverable
set on top of them, ready for review at the next meeting.
Claude