v7.2.0 — Netherlands + Italy focus, English-only, Apify removed

Minor release. Two new search rules + Apify removal.

1. Geography — Netherlands + Italy only

• GEO_COUNTRIES=it,nl (set by run_graph.py / .env) restricts search, ingestion, prefilter AND the Excel export to NL + IT (plus generic remote/Europe-wide). Germany/Spain/UK/France/etc. are dropped.
• Whole-word matching ("roma" inside "Romania" no longer false-matches Italy) + Italian/Dutch provinces & administrative markers ("Provincia di", "Citta Metropolitana", Noord-Holland, ...), so e.g. "Cameri, Provincia di Novara" is correctly kept as Italy.
• Code default is empty (legacy EU-wide) so pipeline tests stay valid — same convention as INTERN_ONLY / ENABLE_*.

2. Language — English-sufficient roles only

• The eligibility language gate now strips HTML before matching and catches the "Italian (fluent)" word-order (e.g. Doctolib's "Italian (fluent) and English (fluent)"), which previously slipped through because HTML tags split the words. A mandatory non-English language => rejected; "a plus / preferred" => kept.
• NON_ENGLISH_AUTO_REJECT is now env-driven (was hardcoded, ignoring .env) and drops JDs written in a non-English language.
• The export re-runs the gate per row, so roles scored before the fix are also dropped from the dashboard.

3. Apify removed

• agents/apify_search.py deleted; no imports remain; requirements.txt & .env.example Apify-free; code comments cleaned; docs updated (+ SUPERSEDED banner on the Phase-2 spec). Sourcing is now Adzuna API + native ATS APIs (Greenhouse/Lever/Ashby/SmartRecruiters) + career-site connectors.

Result: jobs_master.xlsx is NL+IT, intern-only, English-sufficient (Doctolib PM dropped for mandatory Italian; Novara role recovered). 1204 tests passing.

v7.1.3 — Intern-only filter

Patch on v7.1.2.

• Intern-only mode (INTERN_ONLY=1, default): keeps ONLY internship / graduate-program / trainee roles; drops mid/senior roles (Product Manager, Senior PM, Analyst, Associate).
• Applied in the prefilter, job_ingestion, AND the Excel export (so legacy non-intern rows already scored in the DB no longer appear).
• Effect: jobs_master.xlsx went from 447 rows to 94, all internship/graduate.
• 1199 tests passing.

v7.1.2 — Adzuna search + every-run pipeline

Patch on v7.1.1.

• Adzuna source: official free REST API connector (aggregates LinkedIn/Indeed/company-site roles across EU countries — no scraping, no Apify). Enable with ADZUNA_APP_ID / ADZUNA_APP_KEY.
• Search every run: run_graph.py now runs the full pipeline on every run with no opt-in — search -> filter -> score -> documents (>=80%) -> Excel.
• Wired into both the graph and CLI search paths through the existing prefilter/score/export flow.
• 1197 tests passing.

v7.1.1 — Intern-level seniority filter

Patch on v7.1.0.

• Seniority filter: the graph pipeline now drops clearly-senior roles (senior/staff/lead/principal/director/VP/chief) so only internship/junior/entry roles are scored and shown. Intern/graduate signals override (e.g. "Senior … Internship" passes).
• Applied in both the PreFilter node and job_ingestion (the read path used when acquisition is off).
• Effect: scored set ~413 -> ~288; no Senior/Staff/Lead titles in matches.
• 1195 tests passing.

v7.1.0 — Career Intelligence: multi-source search, PreFilter & observability

Built on v7.0.0.

• Multi-source job search: Greenhouse, Lever, Ashby, SmartRecruiters, CareerSiteSource (JSON-LD/RSS); Apify (LinkedIn/Indeed/WTTJ) wired, off until verified actor ids.
• PreFilter LangGraph node + lightweight checkpoint state (~110x smaller).
• Observability: pipeline metrics, /api/v2/metrics/{funnel,scoring,acquisition}, and a /dashboard page.
• Fix: nested candidate_profile.json schema mapping.
• run_graph.py one-command runner.
• VERSION bumped to 7.1.0.
• 1193 tests passing.

v7.0.0 — LangGraph Coexistence Release

v7.0.0 — LangGraph Coexistence Release

Overview

v7.0.0 marks the completion of the LangGraph convergence initiative.

The repository has evolved from a read-only analysis overlay into a replay-safe, effect-capable graph execution engine that can coexist with the legacy pipeline while keeping full replacement as an intentionally deferred architectural option.

Highlights

LangGraph Capability Convergence

• Added replay-safe, side-effect-capable LangGraph execution
• Migrated all six legacy side-effect classes:
  • Acquisition
  • Score persistence
  • Research
  • Document generation
  • Export
  • Tracker updates
• Introduced terminal-stage orchestration for optional stages

Security Hardening

• API-key authentication
• Tenant isolation enforcement
• Tenant-aware LLM cache keys
• Improved rate limiting
• SQLite concurrency improvements
• Worker-pool retry and termination fixes
• CORS configuration support
• Dependency pinning and documentation updates

Reliability Improvements

• Real checkpoint/resume replay testing
• Idempotency guarantees across all side-effect classes
• Atomic document writes using temp-file + fsync + atomic replace
• Extensive replay and security test coverage

Architecture Decision

Adopted operating model:

• Legacy pipeline → Authoritative system of record
• LangGraph → Opt-in execution engine
• Replacement of the legacy orchestrator → Explicitly deferred

Validation

• Test suite: 1,151 passed, 2 skipped
• Replay safety validated through end-to-end checkpoint/resume tests
• All convergence work documented in docs/CONVERGENCE.md

Status

Production-ready for coexistence mode.

Future work will focus on product capabilities and business value creation rather than architectural migration.

Career Intelligence Engine v6.0.0 – Planner-Driven DAG Orchestration Platform

Career Intelligence Engine v6.0.0

🚀 Highlights

• Planner-driven DAG orchestration platform
• Deterministic intent classification
• Dynamic dependency-based routing
• LangGraph checkpoint persistence
• Semantic matching and recommendation engine
• Opportunity intelligence and prioritization framework
• Parallel execution infrastructure
• Production-ready APIs and observability stack
• Extensive documentation and architecture specifications
• 1,101 automated tests passing

🏗️ Architecture Improvements

• ExecutionPlan and Task schemas
• Planner node with dependency resolution
• Dynamic router supporting scoped execution plans
• Supervisor completion tracking via completed_tasks reducer
• Persistent checkpoints and resumable workflows
• Modular agent runtime and orchestration framework

🧪 Quality Metrics

• Total tests: 1,103
• Passed: 1,101
• Skipped: 2
• Failures: 0

📦 Release Status

Production Ready ✅

This release establishes the foundation for planner-driven, dependency-aware, stateful AI workflow orchestration and advances the Career Intelligence Engine into a scalable multi-agent platform.