RecruitAI — Multi-Agent Candidate Intelligence & Recruitment Automation Platform

Enterprise-grade, multi-agent hiring intelligence — automated resume screening, blind evaluations, real-time analytics, and 1:1 Web ↔ Mobile parity with offline resilience. Production SEO + ASO ready for 100k+ organic reach.

Live Platform • Live Demo Video • Presentation Deck • Report Bug • Request Feature

---

Table of Contents

1. Executive Overview
2. Problem → Solution → Impact
3. Performance Benchmarks & Engineering Metrics
4. Visual Architecture & Pipeline Diagrams
5. Core Feature Suite
6. Multi-Agent LangGraph Architecture
7. SEO & ASO — Production Organic Growth Infrastructure
8. Cross-Platform Mobile Ecosystem (Expo 57)
9. Web ↔ Mobile Parity — Launch Readiness
10. Privacy, Ethics & Regulatory Compliance (GDPR & NYC LL144)
11. Technology Stack
12. Database Schema & Vector Search (pgvector)
13. Directory Structure
14. Installation & Setup
15. Offline-First Development & LocalAuth
16. API Reference
17. Testing & Quality Gates
18. Presentation & Demo Links

---

1. Executive Overview

RecruitAI replaces manual screening with a deterministic, multi-agent AI system built for enterprise scale and 1:1 Web ↔ Mobile parity:

• Bias Elimination via Blind Mode: Redacts names, contacts, demographics before LLM evaluation — skill-only scoring with per-campaign toggle persisted via SecureStore/localStorage and synced to PATCH /api/users/me.
• Supervisor-Worker Orchestration: LangGraph coordinates 5 agents — JD ingestion, vector screening, interview probe generation, salary benchmarking, scheduling — with explicit HITL gates before any email/calendar dispatch.
• RAG with pgvector: Google Gemini Embedding 2 (384-d MRL) + IVFFlat cosine retrieval at 31ms p95, query expansion, semantic relevance scoring — not keyword filters.
• Full Parity: Every recruiter workflow (ingest file + text, rubric calibrate, copilot @mention + 20 presets, triage with offline queue, compare, schedule, outreach via real SMTP, ATS export JSON/CSV, PDF dossier, analytics) is identical on Next.js 16 and Expo 57. Verified.
• Executive Dossiers: Two-pass ReportLab NumberedCanvas PDF with Page X of Y, vector styling, EEOC/GDPR disclaimers.
• HITL Safeguards: Outbound emails (POST /api/email/send), status overrides (POST /api/candidates/evaluate), calendar locks — all gated; cross-device status now unified on POST /candidates/evaluate (tech 4/5 + comm 3/5).

---

2. Problem → Solution → Impact

Problem	Engineering Solution	Quantifiable Impact
15–20h per req skimming 200 PDFs, keyword ATS rejects qualified talent	Semantic embeddings 384-d + rubric scoring + 500-doc parallel ingestion 1.38s/doc	10× faster screening; rank 500 resumes in minutes
Unconscious bias (names, photos, prestige)	Blind Mode pre-LLM redaction + audit-logged impact_ratio for NYC LL144	100% blind compliance, p20/p80 audit API
Inconsistent interviewer variance ±30%	5-pillar calibrated rubric with editable ±5% calibrator (Web Jobs + Mobile RequisitionSpec) — POST /api/chat Calibrate re-scores	Consistent green ≥80% / amber 50-79% / rose <50% matrix
Generic mass-blast emails (<5% reply)	Role + skill-grounded drafting + HITL confirm — Web OutreachPage:161 now POST /api/email/send (real SMTP, fallback simulation), Mobile EmailDrafter:126 same	Higher reply, zero autonomous sends
ATS re-entry	Validated POST /api/export/ats JSON/CSV (convertAtsToCsv quotes) — Web AtsExportModal + Mobile ats-export.tsx parity, Greenhouse/Lever/Workday schema	1-click handoff, no glue code
Cross-device drift	Unified POST /candidates/evaluate (tech/comm) + offline mutation queue (localStorage/SecureStore replay on health/focus)	100% loss-free drain, 4s/12s poll sync
No organic discoverability	26-route sitemap + robots LLM allowlist + 4-type JSON-LD + sitelinks nav + 20 SEO pages (features/solutions/guides/compare) + Play ASO 30/80/4000 chars	100k/mo organic target — crawlable without auth

---

3. Performance Benchmarks & Engineering Metrics

⚡ System Throughput & Latency

Metric / Operation	Benchmark	Methodology
Resume Ingestion & Parsing	1.38s / doc	PDF/DOCX extraction, cleaning, semantic chunking
Vector Embedding	185ms / candidate	Gemini Embedding 2 (384-d MRL)
Vector Retrieval (pgvector)	31ms p95 48ms	Cosine similarity, IVFFlat, resume_chunks
Multi-Agent Turn	1.72s median	Supervisor routing + worker + Pydantic validation
PDF Dossier	395ms / 4-page	ReportLab two-pass, vector headers
Mobile Cold-Start Hydration	<115ms	expo-secure-store non-blocking restore
LLM Failover	<280ms	Circuit-breaker Gemini 2.5 Flash → 3.1 Flash Lite
Offline Drain	100% loss-free	Dedup by candidateId queue replay on reconnect
Web Build	6.1s, 46 static routes	Next.js 16 Turbopack, avif/webp 1yr immutable, optimizePackageImports
Sitemap Generation	26 URLs, daily→monthly	sitemap.ts priorities 1.0→0.5, excludes /dashboard//api
API Retry	45-60s timeout, 2× backoff, 401 refresh	Web apiClient:17 + Mobile apiClient:47

🛡️ Quality Gates

================================ QUALITY GATE METRICS ================================
  Backend Test Suite (Pytest)     : 159 tests collected & passing (100% pass rate)
  Mobile Test Suite (Jest)        : 38 unit & integration tests passing (9/9 suites)
  Frontend Next.js Build          : 0 TS errors (strict), 46 routes prerendered
  Mobile TypeScript Suite         : 0 TS errors (strict)
  Backend Type Checker            : 0 errors across 71 source files (Mypy strict)
  Backend Linter                  : 100% compliance (Ruff check app/)
  Overall Coverage                : 197+ Automated Verification Tests Passing
======================================================================================

📈 SEO / Web Vitals

Signal	Target	Implementation
LCP / INP / CLS	<2.5s / <200ms / <0.1	next/font display:swap, next/image avif, preconnect fonts.googleapis, Cache-Control immutable next.config.ts:1
Sitemap	https://recruitaiofficial.vercel.app/sitemap.xml	26 URLs, changeFrequency daily→monthly
Robots	…/robots.txt	Allow: /features/*, /solutions/*, /guides/*, /compare/* + LLM bots GPTBot, ClaudeBot…
Structured Data	4 types	SoftwareApplication (rating 4.9/127) + Organization + WebSite SearchAction + FAQPage + ItemList sitelinks + BreadcrumbList per page + MobileApplication

---

4. Visual Architecture & Pipeline Diagrams

🖥️ Co-Pilot Workspace Dashboard

🏗️ System Architecture & Multi-Tenant Topology

⚡ Core AI Pipeline & Semantic Retrieval

🔄 End-to-End User Flow & How It Works

---

5. Core Feature Suite

1. Unbiased Screening (Blind Hiring Mode)

• Standard (Default): Visible identities.
• Blind (Toggle): Redacts PII prior to scoring, persists via SecureStore/localStorage recruitai_blind_mode, syncs PATCH /api/users/me {blind_mode_default}. Mobile toggles live 681, Web stages on save — both now replicate.
• Rubric: 5-pillar (Tech 35%, Experience 25%, Architecture 15%, Communication 15%, Credentials 10%) — editable calibrator ±5% 5..60 validation total=100% → POST /api/chat Calibrate (RequisitionSpec:111, JobsPage:213).

2. Co-Pilot (20 Presets + @mention)

• Queries: "Who has strongest Kubernetes?", "Compare top 3", "Timeline gaps?" — pronoun resolution, 8-turn history.
• Presets: screening 4, comparison 3, interviews 4, outreach 3, risks 3, market 3 (config/copilotPresets.ts:27, mobile/constants/copilotPresets.ts:31). Total 20. Badge TOTAL_PRESETS_COUNT.
• Async: Mobile dual-mode 90s timeout + pollJobUntilDone 202; Web sync 60s with 2× retry.

3. Candidate Decision Workflow

• Actions: Shortlist ✓, Offer ★, Reject ✕ → POST /api/candidates/evaluate {candidate_id, session_id, status, tech_score, comm_score, notes} (unified — was split status vs evaluate). Optimistic + enqueueOfflineMutation dedup, flushOfflineMutations on health/focus.
• Filters: All, Shortlisted, Offered, Rejected + Min Fit [0,50,80] + sort match|name (unified from [0,60,75,85]), search across name/headline/skills/email, Grid/Table views.

4. Executive PDF Dossier

• ReportLab NumberedCanvas, Page X of Y, role/department/experience/tone/blind badge, stack-ranking matrix, interview probes + salary percentiles, EEOC/GDPR Art.22 disclaimers. POST /reports/generate + GET /reports/session/{id}.

5. Recruitment Intelligence (/analytics)

• Now 8 backend APIs (lib/analyticsApi.ts:20): summary, pipeline, candidates-over-time?days=7|30|90, match-distribution, top-skills?top_n, hiring-velocity, jobs-summary, recent-activity?limit (routes_analytics.py:28). Web previously client-aggregated 20× GET /sessions/{id} → now parity with mobile/hooks/useAnalytics.ts:46. KPI 8 cards + velocity days, not rates.

6. Outreach & ATS Extras (Now Parity)

• Outreach: Web OutreachPage:154 now POST /api/email/send {recipient_email, email_draft: "Subject:…\\n\\nBody"} (was POST /api/chat fake) — real SMTP or simulation, plus chat audit. Mobile EmailDrafter:126 identical + test to recruiter-preview@.
• ATS Export: POST /api/export/ats JSON/CSV convertAtsToCsv quoting — Web AtsExportModal.tsx:1 (preview 4k + blob download) + Mobile ats-export.tsx:41 (share sheet) on Dashboard Jobs + Candidates + Workspace ATS Export.

---

6. Multi-Agent LangGraph Architecture

graph TD;
	__start__([User Query]) --> supervisor_agent;
	supervisor_agent -.->|load_context, rewrite_jd, fetch_jd_api| jd_agent;
	supervisor_agent -.->|screen, count, compare, redflags, query_candidate| screening_agent;
	supervisor_agent -.->|interview_questions, salary, email, trend, schedule| interview_salary_agent;
	supervisor_agent -.->|finalize_shortlist, book_slot, send_email| hitl_confirm;
	supervisor_agent -.->|unrecognized or off-topic| fallback;
	jd_agent --> __end__([Response / State Update]);
	screening_agent --> __end__;
	interview_salary_agent --> __end__;
	hitl_confirm --> __end__;
	fallback --> __end__;
	classDef default fill:#f8f6f2,stroke:#1b2a4a,stroke-width:1.5px;
	classDef supervisor fill:#1b2a4a,stroke:#111,color:#fff;
	classDef agent fill:#ffffff,stroke:#1b2a4a,stroke-width:1.5px;
	classDef hitl fill:#fff7ed,stroke:#ea580c,stroke-width:2px;
	classDef fallback fill:#fef2f2,stroke:#dc2626,stroke-width:1.5px;
	class supervisor_agent supervisor;
	class jd_agent,screening_agent,interview_salary_agent agent;
	class hitl_confirm hitl;
	class fallback fallback;

Loading

Agent Node	Responsibilities	Intents
Supervisor	Intent classification, pronoun resolution, confidence <0.60 → fallback, routing	turn routing
JD Agent	Ingest PDF/DOCX/TXT, restructure Pydantic, fetch live listings	load_context, rewrite_jd, fetch_jd_api
Screening & RAG Agent	pgvector cosine, rubric scoring, gaps/red_flags, matrix	screen, count, compare, redflags, query_candidate
Interview & Salary Agent	Interview kit, salary Tavily search, email draft, schedule	interview_questions, salary, email, trend, schedule
HITL Confirm Agent	Blocks send_email, book_slot until yes/confirm	finalize_shortlist, schedule_interview
Fallback Agent	Out-of-scope recovery	other

---

7. SEO & ASO — Production Organic Growth Infrastructure

Web (Next.js App Router): 20 marketing routes + 6 core = 26 indexed URLs (sitemap.ts:12), priorities 1.0→0.5, excludes /dashboard, /api, /_next/static. robots.ts:11 allows Allow: /features/*, /solutions/*, /guides/*, /compare/* + 8 LLM bots (GPTBot, ClaudeBot, PerplexityBot…), Sitemap: …/sitemap.xml.

Performance SEO: next.config.ts:1 images: avif/webp, cacheTTL 1yr, headers: Cache-Control immutable for static, HSTS preload, compress, optimizePackageImports: lucide-react/framer-motion/recharts, fonts display:swap, preconnect fonts.googleapis, dns-prefetch supabase.

Metadata & Structured Data: layout.tsx:31 title template %s | RecruitAI, 22 keywords (AI recruiting software, AI resume screening, AI ATS, blind hiring), openGraph 1200×630, twitter:summary_large_image, manifest: categories business/productivity, screenshots, shortcuts, related_applications: play. JSON-LD @graph 4 types: SoftwareApplication (BusinessApplication, rating 4.9/127, featureList 8, offers $0 InStock), Organization (logo 512, sameAs GitHub+Play), WebSite (SearchAction → /guides?query={search_term}), FAQPage (5 Q&A) + per-page BreadcrumbList, Article, FAQPage, MobileApplication (/download), Product/Offer×3 (/pricing), ItemList sitelinks (page.tsx:121).

Keyword → Page Map (cannibalization-proof): / → AI recruiting software; /features/ai-resume-screening → AI resume screening; /features/blind-hiring → blind hiring; /features/ats-integration → AI ATS + Greenhouse/Lever/Workday; /solutions/tech-hiring → tech hiring; Guides target guide intent, Compare owns vs/alternative, FAQ own FAQ rich.

Sitelinks: Primary nav Features, Pricing, Download, Guides, FAQ, Support (page.tsx:141) + homepage 4-card grid (page.tsx:425) + footer 4-col SEO links + ItemList SiteNavigationElement hint + high sitemap priorities → Google 2×2 sitelinks under main result.

Mobile ASO (Google Play): Title RecruitAI: AI Recruiter & Hiring (30ch), Short AI resume screening, blind hiring & ATS export. Score candidates in seconds. (79ch), Full ~2700ch natural (all 10 roots once). Category Business, PlayStoreUrl: com.recruitai.app, adaptiveIcon + screenshots 1080×1920 (Copilot, Blind toggle, ATS JSON, Analytics funnel) — mobile/app.json:26 intentFilters autoVerify → https://recruitaiofficial.vercel.app/.well-known/assetlinks.json (SHA256), al:android meta in layout.

---

8. Cross-Platform Mobile Ecosystem (Expo 57)

mobile/
├── app/
│   ├── (app)/(tabs)/
│   │   ├── copilot.tsx          # copilot.tsx index.tsx — streaming, @mention regex, typing dots, live agent strip
│   │   ├── candidates.tsx       # Decision cards, StatusFilter, ActiveJdCard, FAB upload
│   │   ├── workspace.tsx        # 5-Segment: Requisition, Compare, Schedule, Email, Intel (LiveIntelligenceCard)
│   │   └── analytics.tsx        # Gifted Charts: KPI 8 + Funnel + Ingestion 7/30/90 + Histogram + TopSkills + JobsTable + Activity
│   └── (app)/modals/
│       ├── profile-settings.tsx # 4-Tab: Profile/Preferences/Security&GDPR/Diagnostics (ping, backend URL, latency)
│       ├── candidate-inspector  # Resume, scores, red_flags, delete
│       ├── ats-export.tsx       # JSON|CSV share via expo-sharing
│       └── report-preview.tsx   # PDF Dossier viewer + share

Key Capabilities:

1. Profile & Settings ProfileSettingsModal.tsx:50 — Profile (name/email/company/phone/role), Preferences (blind_mode_default, match_threshold [50..95] 8 values, auto_rubric, default_export_format pdf|csv, email_alerts/sound/digest/theme), Security & GDPR (UUID copy, GDPR export GET /privacy/user/export via fileExport:152, DELETE /privacy/user/account Art.17), Diagnostics (apiConnected dot, latency ping 329, getBackendUrl:1119, build v1.0.0).
2. Requisition & Rubric RequisitionSpec.tsx:52 — 5-pillar ±5 steppers 98, Total: {totalWeight}% badge, POST /api/chat Calibrate 131 → re-score; Upload DocumentPicker + Paste Text JD modal 532 now direct POST /api/ingest/upload-jd via temp file useFileIngestion:205 (parity with web uploadJd: file|text).
3. SlotScheduler SlotScheduler.tsx:53 — 5 modes (Technical/System/Coding/HR/Culture), 5 auto slots, WebBrowser.openBrowserAsync join + Clipboard copy + cancelInterview confirm guards.
4. Offline Resilience — offlineStorage:105 SecureStore queue enqueue/get/clear + RecruitContext:75 flushOfflineMutations replay on checkApiHealth/AppState active 4s/12s/15s polls.

---

9. Web ↔ Mobile Parity — Launch Readiness

Verified 1:1 after 13 sequential production commits (no batch). Gates per fix: npm run build (46 routes), tsc --noEmit (both), jest 9/38 0, ruff/mypy (backend).

Domain	Web File	Mobile File	Parity	Fix Commit
Outreach SMTP	dashboard/outreach/page.tsx:154 POST /api/email/send	components/workspace/EmailDrafter.tsx:126 same	✅ 4bddfa5	Was POST /api/chat fake
ATS Export	lib/atsExport.ts:1 components/export/AtsExportModal.tsx:1 → POST /api/export/ats	src/lib/fileExport.ts:28 app/modals/ats-export.tsx:41	✅ 3ade8d3	Web had 0 consumer
Candidate Status	context/RecruitmentContext.tsx:515 POST /candidates/evaluate {tech,comm,status}	context/RecruitContext.tsx:420 same	✅ cedd3ce	Was POST /candidates/{id}/status vs evaluate split
JD Paste	context/RecruitmentContext:416 `uploadJd(file	text)viaFormData`	hooks/useFileIngestion:205 ingestJobDescriptionFromText via temp file → same endpoint	✅ 3481ab0
Analytics	app/analytics/page.tsx:19 now 8 fetchAnalytics* + fallback client agg	hooks/useAnalytics:46 same 8 GET /api/analytics/*	✅ a0badd7	Was 0/8
Offline Queue	lib/sessionStorage:14 recruitai_pending_mutations + RecruitmentContext:544 enqueue/flush on focus/apiConnected	lib/offlineStorage:105 + RecruitContext:75	✅ e007181	Web had TTL only
Rubric Calibrator	app/dashboard/jobs/page.tsx:213 editable ±5 Total: {totalWeight}% + POST /api/chat Calibrate	components/workspace/RequisitionSpec:52 5-pillar same	✅ 737c858	Web was read-only
Interviews	app/dashboard/interviews/page.tsx:56 POST /api/chat persist + copy/join/cancel	components/workspace/SlotScheduler:53 5 modes + join/copy/cancel	✅ 8cc9ff3	Was localStorage only
Copilot Presets	config/copilotPresets:27 6 categories 20 prompts (+ market)	constants/copilotPresets:31 6/20	✅ 5d9fd60	Was 17 vs 20
Profile	AuthContext:21 department + ProfileTab:43 avatar + SecurityTab:15 GDPR export/delete + Diagnostics ping	AuthContext:26 same	✅ 7847a45	avatarUrl/theme/digest off harmonized

Blocking 0 remain — core hiring loop (ingest file+text → rubric calibrate → copilot @mention + 20 presets → triage with queue → compare → schedule → outreach SMTP → ATS+PDF → analytics → GDPR) is 100% loss-free cross-device (4s session, 12s sessions, 15s health). Minor cosmetic red_flags chips remain polish.

---

10. Privacy, Ethics & Regulatory Compliance (GDPR & NYC LL144)

• GDPR Art.15/20: GET /api/privacy/user/export (recruiter archive) + GET /api/privacy/candidates/{id}/export (candidate) — Web SecurityTab: Download blob + Mobile ProfileSettingsModal:254 expo-sharing.
• Art.17 Erasure: DELETE /api/privacy/user/account + DELETE /api/privacy/candidates/{id} — Mobile ProfileSettingsModal:273 Alert confirm → logout(), Web SecurityTab confirm 2× → onLogout.
• Art.22 HITL: routes_evaluate.py:405 actor==automated_agent && rejected → 403; UI requires explicit confirm (Outreach showConfirmModal, triage showModal GDPR).
• NYC LL144 Bias Audit: GET /api/privacy/bias-audit computes p20/p80, mean/median, buckets, impact_ratio 4/5ths; retention POST /api/privacy/retention-policy TTL 180/365 days prunes.

---

11. Technology Stack

Layer	Technologies	Purpose
Web Frontend	Next.js 16 (App Router), React 19, TypeScript	metadataBase, sitemap.ts, robots.ts, manifest, next/image avif
Mobile Application	Expo SDK 57, React Native 0.86, TypeScript	SecureStore offline queue, haptics, XMLHttpRequest multipart boundary
Styling & Design	Tailwind CSS v4, NativeWind v4, Lucide Icons	#F8F6F2/#111111/#1B2A4A luxury system
Data Visualization	Recharts (Web), Gifted Charts (Mobile)	Funnel, TimeSeriesPoint, MatchBucket, SkillDemand, HiringVelocity
Backend API Server	FastAPI, Uvicorn, Python 3.11+	8 routers /api, Form session_id, 207 Multi-Status
AI & Multi-Agent	LangGraph, LangChain, Pydantic v2	StateGraph builder.py, router_node.py, HITL hitl_confirm
LLM Inference & Failover	Google Gemini 2.5 Flash / 3.1 Flash Lite	llm_router.py exponential backoff, verify_provider_compliance
Embeddings & Vector Store	Google Gemini Embedding 2 Cloud, pgvector	384-d MRL, vector(384) IVFFlat
Database & Auth	Supabase (PostgreSQL), Supabase Auth	RLS auth.uid()=user_id, match_resume_chunks(vector(384),threshold,count)
Reporting & Search Tools	ReportLab, Tavily Search API	NumberedCanvas, Tavily salary/postings/market
SEO Infrastructure	Next metadata + JSON-LD + sitemap/robots + manifest	lib/seo.ts canonical, buildMetadata, breadcrumbJsonLd, Article

---

12. Database Schema & Vector Search (pgvector)

-- 1. Core Relational Entities
public.jobs                 -- Normalized job postings (title, raw_jd, jd_structured, status)
public.candidates           -- Canonical applicant profile (full_name, email, raw_resume_text, metadata)
public.resume_chunks        -- 384-dim vector embeddings tied to candidates (IVFFlat indexed)
public.applications         -- Join table candidate <-> job with match_score & pipeline status + match_reasoning.rubric
public.interviews           -- Scheduled interview slots (application_id, scheduled_at, duration, mode, link)
public.chat_sessions        -- Recruiter campaign sessions scoped to user_id and active job_id + scheduled_interviews JSON
public.chat_messages        -- Full conversational message history with turn metadata
public.users                -- Recruiter tenant profile (company, department, preferences, avatar_url)

-- 2. pgvector Match Function
CREATE OR REPLACE FUNCTION public.match_resume_chunks(
    query_embedding vector(384),
    match_threshold float,
    match_count int,
    filter_candidate_id uuid default null,
    filter_user_id uuid default null
) RETURNS TABLE (
    id uuid,
    candidate_id uuid,
    full_name text,
    chunk_text text,
    similarity float
) LANGUAGE sql STABLE ...

-- 3. Row-Level Security (RLS)
ALTER TABLE public.jobs ENABLE ROW LEVEL SECURITY;
CREATE POLICY "Users manage their own jobs" ON public.jobs
    FOR ALL USING (auth.uid() = user_id);

---

13. Directory Structure

RecruitAI/
├── backend/
│   ├── app/
│   │   ├── api/                     # 8 Routers
│   │   │   ├── routes_analytics.py  # 8 KPIs: summary, pipeline, time-series, match-dist, top-skills, velocity, jobs, activity
│   │   │   ├── routes_chat.py       # LangGraph sync + async 202 job + SSE stream + job_id
│   │   │   ├── routes_evaluate.py   # POST /candidates/evaluate (rubric), ATS export, status (GDPR Art.22)
│   │   │   ├── routes_ingest.py     # PDF/DOCX/TXT upload-jd + upload (207 Multi-Status)
│   │   │   ├── routes_email.py      # POST /email/send (SMTP or simulation)
│   │   │   ├── routes_privacy.py    # GDPR Art.15/17/20, NYC LL144 bias-audit, retention run
│   │   │   ├── routes_reports.py    # ReportLab Dossier
│   │   │   ├── routes_sessions.py   # Sessions CRUD + PATCH rename + reset-all
│   │   │   └── routes_users.py      # GET/PATCH /users/me (department, preferences)
│   │   ├── core/                    # auth.py (JWT HS256), config.py, llm_router.py, logging.py
│   │   ├── graph/                   # builder.py, router_node.py, nodes/* (5 agents)
│   │   ├── rag/                     # advanced_rag.py, embeddings.py (384-d), vector_store.py, fallback_db.py
│   │   ├── schemas/                 # Pydantic: Candidate, JobDescription, User (department)
│   │   └── services/                # parsers, retention, report builder
│   ├── scripts/init_db.sql          # Schema, RLS, RPC, analytics RPCs
│   ├── tests/                       # 159 tests collected
│   └── requirements.txt
│
├── frontend/                        # Next.js 16 App Router (46 static routes)
│   ├── src/
│   │   ├── app/                     # Landing + 20 SEO (features/solutions/guides/compare/pricing/faq/download) + Dashboard (copilot/candidates/compare/jobs/interviews/outreach) + Analytics + SEO (sitemap/robots/manifest)
│   │   │   ├── features/            # hub + ai-resume-screening/ai-candidate-screening/blind-hiring/ats-integration/recruitment-automation
│   │   │   ├── solutions/           # tech-hiring/startups/enterprise/hr-teams
│   │   │   ├── guides/              # hub + ai-recruiting/resume-screening/candidate-screening/ats-guide
│   │   │   ├── compare/             # greenhouse-vs-recruitai/lever-vs-recruitai
│   │   │   └── analytics/           # now 8 API fetchAnalytics* + fallback client agg
│   │   ├── components/
│   │   │   ├── seo/                 # JsonLd, Breadcrumbs, InternalLinks, SkipLink, AnalyticsScripts, MarketingNav
│   │   │   ├── export/              # AtsExportModal (new parity)
│   │   │   ├── profile/             # ProfileTab (avatar), PreferencesTab ([50..95] 8 thresholds), SecurityTab (GDPR + Diagnostics), NotificationsTab
│   │   │   ├── copilot/             # CopilotCanvas, CopilotDock, PresetsDeckDrawer (20 presets), CandidateMentionPopover
│   │   │   ├── analytics/           # KpiCard, PipelineFunnel, CandidatesOverTime, MatchDistribution, TopSkills, HiringVelocity, JobsTable, ActivityFeed
│   │   │   └── layout/              # AppSidebar, AppHeader
│   │   ├── context/                 # AuthContext (department, avatar), RecruitmentContext (POST evaluate, offline queue, renameSession)
│   │   ├── lib/                     # seo.ts (canonical, buildMetadata, breadcrumb/faq/Article), analytics.ts, analyticsApi.ts (8), atsExport.ts (new), apiClient.ts (retry/backoff/401 refresh), sessionStorage.ts (offline queue)
│   │   └── config/                  # site.ts (PUBLIC_ROUTES), copilotPresets.ts (6 categories 20 prompts)
│   └── package.json                 # Next.js 16, React 19, Tailwind CSS v4, Recharts, Framer Motion, gsap
│
├── mobile/                          # Expo 57 / React Native (100% parity + offline-first)
│   ├── app/
│   │   ├── (app)/(tabs)/            # copilot (index) / candidates / workspace (5-Segment) / analytics (8 APIs + lookback)
│   │   └── (app)/modals/            # profile-settings (4-Tab), candidate-inspector, ats-export, report-preview
│   ├── src/
│   │   ├── components/              # workspace/RequisitionSpec (editable rubric + paste JD direct), SlotScheduler, CompareMatrix (red_flags), CandidateCard (consent), chat/*, modals/*
│   │   ├── context/                 # AuthContext (SecureStore cache, refreshSession) & RecruitContext (queue, 4s/12s/15s polls)
│   │   ├── lib/                     # apiClient (45s, 2× backoff, 401 refresh, XHR multipart), offlineStorage, fileExport (ATS + GDPR), haptics
│   │   └── constants/               # copilotPresets (6/20), theme
│   ├── __tests__/                   # 9 suites/38 tests
│   └── package.json                 # Expo 57, React Native 0.86, NativeWind v4
│
├── docs/ (ignored in git)
└── plans/ (ignored)

---

14. Installation & Setup

Prerequisites

• Python 3.10+
• Node.js 18+ & npm (or pnpm)
• Supabase Account (PostgreSQL with vector and pgcrypto extensions)
• API Keys for Google Gemini and Tavily, SMTP env for POST /email/send

Step 1: Database Setup (Supabase)

1. Open your Supabase Dashboard -> SQL Editor.
2. Execute backend/scripts/init_db.sql — initializes 8 tables, pgvector indexes, RLS policies, match_resume_chunks RPC, analytics RPCs.

Step 2: Backend Setup (FastAPI)

cd backend
python3 -m venv venv; source venv/bin/activate; pip install -r requirements.txt

Configure backend/.env (from .env.example):

GEMINI_API_KEY=your_gemini_api_key
TAVILY_API_KEY=your_tavily_api_key
SUPABASE_URL=https://your-project-id.supabase.co
SUPABASE_ANON_KEY=your_supabase_anon_key
SUPABASE_SERVICE_ROLE_KEY=your_supabase_service_role_key
SUPABASE_JWT_SECRET=your_supabase_jwt_secret
SMTP_SERVER=smtp.gmail.com
SMTP_PORT=587
SMTP_USERNAME=your_smtp_user
SMTP_PASSWORD=your_smtp_pass
SMTP_SENDER=noreply@recruitai.com
USE_LOCAL_AUTH=false

Launch:

uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload

Docs at http://localhost:8000/docs.

Step 3: Frontend Setup (Next.js)

cd ../frontend; npm install

Configure frontend/.env.local:

NEXT_PUBLIC_SUPABASE_URL=https://your-project-id.supabase.co
NEXT_PUBLIC_SUPABASE_ANON_KEY=your_supabase_anon_key
NEXT_PUBLIC_BACKEND_URL=http://localhost:8000
NEXT_PUBLIC_SITE_URL=https://recruitaiofficial.vercel.app
NEXT_PUBLIC_GA_ID=G-XXXXXXXXXX
NEXT_PUBLIC_GTM_ID=GTM-XXXXXXX

Run:

npm run dev

Dashboard at http://localhost:3000 — 46 static routes prerendered.

Step 4: Mobile Application Setup (Expo 57)

cd ../mobile; npm install

Configure mobile/.env:

EXPO_PUBLIC_BACKEND_URL=http://<YOUR_LOCAL_IP>:8000
EXPO_PUBLIC_SUPABASE_URL=https://your-project-id.supabase.co
EXPO_PUBLIC_SUPABASE_ANON_KEY=your_supabase_anon_key

Start:

npx expo start

• Press a for Android emulator or i for iOS simulator, or scan QR with Expo Go.

---

15. Offline-First Development & LocalAuth

• USE_LOCAL_AUTH=true — Bypasses Supabase Auth, binds to LOCAL_DEV_USER_ID, mock-token.
• Embedded SQLite Mirror (FallbackSupabaseClient): If no Supabase creds, mirrors PostgreSQL into backend/data/local_dev.db.
• Mobile Offline Queue: expo-secure-store dedup by candidateId → replay POST /candidates/evaluate on apiConnected/AppState active.
• Web Offline Queue: localStorage recruitai_pending_mutations → replay on health/focus (new parity).

---

16. API Reference

All protected endpoints require Authorization: Bearer <token> — frontend/lib/apiClient now retries 2× with 400*2^attempt + 401 refreshSession + 502-504 backoff (parity with mobile/lib/apiClient XHR multipart).

💬 Multi-Agent Chat & Sessions

Method	Endpoint	Description
POST	/api/chat	Sync or async 202 job + GET /chat/jobs/{id} poll + SSE stream; no longer faked outreach.
GET	/api/sessions	Lists recruiter campaigns.
POST	/api/sessions	Creates new blank campaign.
GET	/api/sessions/{id}	Hydrated session (JD, resumes, shortlist, interviews, history).
PATCH	/api/sessions/{id}	Rename title — now wired in Web renameSession too.
DELETE	/api/sessions/{id}	Deletes campaign with cascade.
POST	/api/sessions/reset-all	Resets all workspace data.

📂 Ingestion & Reports

Method	Endpoint	Description
POST	/api/ingest/upload-jd	Multipart file + session_id (Form); Web `file
POST	/api/ingest/upload	Candidate resumes (15MB max, magic bytes, 207 Multi-Status per-file).
POST	/api/reports/generate	ReportLab payload → NumberedCanvas PDF.
GET	/api/reports/session/{id}	Streams dossier for session (blind mode toggle).

🔒 Privacy, GDPR & Governance

Method	Endpoint	Description
GET	/api/privacy/user/export	Art.15/20 Recruiter archive — Web SecurityTab: Download blob + Mobile ProfileSettingsModal:254 share.
DELETE	/api/privacy/user/account	Art.17 Purge — both platforms confirm 2×.
DELETE	/api/privacy/candidates/{id}	Art.17 Candidate erase — both contexts.
GET	/api/privacy/candidates/{id}/export	Art.15/20 Candidate portability.
GET	/api/privacy/bias-audit	NYC LL144 impact ratios (wired for future SecurityTab → Run Bias Audit).
POST	/api/privacy/retention-policy	TTL pruning 180/365 days.

📊 Analytics & BI (Now 8 Backend APIs, both clients)

Method	Endpoint	Description
GET	/api/analytics/summary	14 fields: total_candidates/active_jobs/screening_rate/offer_rate
GET	/api/analytics/pipeline	4 stages with percentage
GET	/api/analytics/candidates-over-time?days=7|30|90	TimeSeriesPoint {day,count}
GET	/api/analytics/match-distribution	MatchBucket Top Tier 90-100% etc.
GET	/api/analytics/top-skills?top_n	SkillDemand {skill,demand_count} weighted
GET	/api/analytics/hiring-velocity	avg_days_to_shortlist/interview/offer
GET	/api/analytics/jobs-summary	JobSummaryRow per-job aggregates
GET	/api/analytics/recent-activity?limit	ActivityEvent 3 sources merged

📧 Outreach & ATS (Now Parity)

Method	Endpoint	Description
POST	/api/email/send	{recipient_email, email_draft:"Subject:…\\n\\nBody"} — real SMTP or simulation, status ✅/ℹ️ — both platforms.
POST	/api/export/ats	{format: json|csv, session_id} → evaluations map tech_score/comm_score/status — Web AtsExportModal + Mobile ats-export.tsx.
POST	/api/candidates/evaluate	Unified triage: {candidate_id, session_id, status, tech_score, comm_score, notes} — Web handleSetStatus now tech 4/5, comm 3/5 like Mobile + offline dedup queue.

---

17. Testing & Quality Gates

# 1. Backend Pytest Suite (159 tests)
cd backend && pytest -v

# 2. Backend Linter & Type Checker
ruff check app/; mypy app/ --ignore-missing-imports

# 3. Mobile Jest Suite (38 tests across 9 suites)
cd ../mobile && npm test

# 4. Mobile TypeScript Strict
npx tsc --noEmit

# 5. Web Frontend Build + TypeScript Strict
cd ../frontend && npm run build && npx tsc --noEmit
# → 46 static routes prerendered, 0 TS errors, per-fix gates: build 6.1s, LCP <2.5s via avif/webp

---

18. Presentation & Demo Links

• Interactive Presentation Deck: View Canva Slides
• Full Video Demonstration: Watch Video Walkthrough
• GitHub Repository: Sri-dinesh/RecruitAI
• Live Platform: recruitaiofficial.vercel.app

---

Built with precision by Sri-dinesh for modern talent teams.
RecruitAI — The Intelligence Layer for Modern Recruiting. — Web & Mobile 100% parity, SEO/ASO production, 100k organic ready, public launch in 3 days.