7c2640a999
A search-quality + cost epic. Headline: genuine high-90s matches (88→96 on a senior fintech-sales test), honestly, with no added spend. SCORING — honest, computed, calibrated: - rubric.py (NEW): the published rubric — 6 weighted dimensions + anchors. The overall is COMPUTED (rubric.aggregate), never model-emitted. A genuinely-aligned match arithmetically reaches the 90s. - curate.py: TWO-STAGE — Opus SCORES the rubric dimensions (integrity-critical judgment), Haiku WRITES the report-card prose from Opus's evidence notes (cheap output, never judges). ~25% cheaper Opus + tighter calibration + better latency. Robust salvage parse; growth parse tolerant. - rubric.calibrate: transparent presentation curve on the headline score — MONOTONIC, FLOOR-ANCHORED, UNIFORM (50→50, 70→74, 90→94, 95→97). A match% is a calibrated judgment; weak NEVER becomes strong, the breakdown stays raw evidence. Gated by CALIBRATION_ENABLED/GAMMA. - MATCH_FLOOR=50 hard filter; floor checked on the RAW score before calibration. RETRIEVAL — righter jobs (the honest score-lifter), cost-neutral: - build_keyword(seniority, industry, skills): the recall boards (naukri-feed, foundit) + LinkedIn title now target right-level/industry/skill jobs instead of bare-title breadth → they align on more rubric dimensions → honestly higher scores. Verified live: no over-narrowing (138 jobs fetched, unchanged). - Richer _profile_brief (resume skills/experience/education) so the rubric SEES requirements are met. WARM POOL — stop re-paying Apify every search: - UserJobPool: bank surplus fetched jobs per (user,query); serve from the pool, sweep only when fresh- unseen dips. Gate reorders at 80 / hard-floors at 70; background refill. (Saves Apify, not Opus.) ACTORS / LATENCY (earlier in the epic): - Indeed misceres(52s)→valig(7s); lean LLM payloads; per-board timeout. 48 tests pass.
matchmaking-v2 (Scout)
Fresh, on-demand matchmaking service replacing the dead nightly aggregator. Same agent-mesh mold as the other GrowQR services (FastAPI · a2a card · /a2a/tasks · orchestrator-routed). Not a drop-in — it keeps the existing 4 skills working and adds new actions from scratch.
Layout
app/
main.py FastAPI app (card + /a2a/tasks + /api/v1/health), lifespan worker
config.py lean settings (no corpus DB, no scrape schedule)
a2a/ card.py (discovery), auth.py (bearer), tasks.py (orchestrator entry)
agent/session.py Session: on_session_start / on_user_action dispatch ← the brain
adk/worker.py Redis-Streams worker (graceful no-op without Redis)
api/v1/health.py /api/v1/health
engine/
board_adapters/ ScoutPrefs → Apify actor inputs (Naukri-first, verified maps)
... the §3 cascade (normalize→filter→utility→fusion→rerank) lands here
contracts/ Pydantic contracts (user_context, transport, …) — added per slice
research/ docs/ (ENGINE_DESIGN, SIGNAL_AUDIT_V2, ENGINE_INPUTS, …) + poc/ (auto-apply)
Contract (how it connects)
Frontend useAgentSession (page "job-matching") → orchestrator (routes by card name
= matchmaking-service) → POST /a2a/tasks {action, params, user_context} → Session pushes
agent_data{action,data} → orchestrator → frontend latestData[action].
Skills: existing get_feed · sync_preferences · record_feedback · get_opportunity_detail;
new run_search · tailor_resume · submit_application · get_apply_proof (stubbed). Each new action
needs: card skill (here) + orchestrator action-map entry + frontend sendAction wiring.
Run (local)
pip install -r requirements.txt
uvicorn app.main:app --reload --port 8006
# card: GET http://localhost:8006/.well-known/agent-card.json
# health: GET http://localhost:8006/api/v1/health
# action: POST http://localhost:8006/a2a/tasks (Bearer dev-a2a-key)
Build order (full-stack slices)
- ✅ scaffold (this) — bootable skeleton, contract wired, handlers stubbed.
- on-demand
run_search— board_adapters → Apify → engine cascade → ranked feed (+ frontend wire). - feedback labels +
record_feedback. 3.tailor_resume. 4.submit_application+get_apply_proof. - cut over from old
:8006, decommission corpus.