feat: add LiteLLM support and auto-apply cron job

- Add get_llm_base_url() and get_llm_model() helper functions
- Update call_ollama_inline to route to LiteLLM when LLM_PROVIDER=litellm
- Add auto-apply cron task for background job matching
- Auto-apply matches job seekers to new jobs based on skills
- Generate cover letters via LiteLLM for each application

.mavis/plans/e2e-full-test.yaml

@@ -0,0 +1,328 @@
+version: 1
+plan:
+  name: nxtgauge-full-e2e-test
+  max_concurrency: 6
+  max_consecutive_failures: 3
+  max_cycles: 10
+  auto_reject_retries: 1
+tasks:
+  - id: track-1-professions
+    title: Test all 10 profession services
+    prompt: |
+      You are testing the Nxtgauge backend API. All tests are via direct HTTP API calls.
+
+      BASE_URL: http://localhost:9100
+      GATEWAY is at :9100 (routes to all services)
+      Individual service ports:
+        photographers: 9107, makeup_artists: 9109, tutors: 9108, developers: 9110,
+        video_editors: 9111, graphic_designers: 9112, social_media_managers: 9113,
+        fitness_trainers: 9114, catering_services: 9115, ugc_content_creators: 9117
+      Test accounts (use test_mode=true to bypass OTP):
+        Professional: bug2_test_1781086381@test.com / Test1234! (PHOTOGRAPHER role active)
+        Customer: cu_1781041378@test.com / Test1234! (CUSTOMER role active)
+      Open lead ID: ba18c522-b3cd-43a7-a269-90cf981c173e (PHOTOGRAPHER, status OPEN)
+
+      FOR EACH of the 10 profession services below, test:
+      1. Health check: GET http://localhost:<PORT>/health or via gateway
+      2. Login to get a fresh token (use test_mode=true register, then login)
+      3. GET /api/<profession>/profile/me — does profile endpoint work?
+      4. GET /api/<profession>/marketplace — does marketplace browse work?
+      5. Send a lead request via POST /api/<profession>/leads/request with body {"requirement_id":"ba18c522-b3cd-43a7-a269-90cf981c173e","message":"interested"}
+
+      Professions to test:
+      - photographers (9107)
+      - makeup_artists (9109) — route is /api/makeup-artists
+      - tutors (9108)
+      - developers (9110) — route is /api/developers
+      - video_editors (9111) — route is /api/video-editors
+      - graphic_designers (9112) — route is /api/graphic-designers
+      - social_media_managers (9113) — route is /api/social-media-managers
+      - fitness_trainers (9114) — route is /api/fitness-trainers
+      - catering_services (9115) — route is /api/catering-services
+      - ugc_content_creators (9117) — route is /api/ugc-content-creators
+
+      NOTE: The "professional" user (bug2_test_1781086381@test.com) only has PHOTOGRAPHER role. For other professions, register a fresh user, register the role, switch to that role, then test.
+
+      Write results to /Users/ashwin/workspace/e2e-results/track-1-professions.md with:
+      - For each profession: service port, health status, profile result, marketplace result, lead_request result, any errors
+      - Summary of which services are fully functional vs partially working vs broken
+    assigned_to: general
+    verify_skip_reason: "Downstream compilation task reads all result files directly; no independent verification needed for raw API test output."
+    max_retries: 1
+
+  - id: track-2-leads
+    title: Test leads service and lead request accept/reject flow
+    prompt: |
+      You are testing the Nxtgauge backend API. All tests are via direct HTTP API calls.
+
+      Gateway: http://localhost:9100
+      Leads service: http://localhost:9118 (or via gateway /api/leads)
+      Customers service: http://localhost:9105
+
+      Test accounts (use test_mode=true):
+        Customer: cu_1781041378@test.com / Test1234!
+        Photographer: bug2_test_1781086381@test.com / Test1234! (has PHOTOGRAPHER approved profile)
+
+      Test the following flows:
+
+      1. LEADS SERVICE CRUD:
+         - GET /api/leads — list all leads (or http://localhost:9118/leads)
+         - POST /api/leads — create a new lead {"title":"Test Lead 2","description":"API test","location":"Delhi","profession_key":"PHOTOGRAPHER"}
+         - GET /api/leads/:id — get specific lead
+         - PATCH /api/leads/:id — update the lead title
+
+      2. CUSTOMER VIEW REQUESTS (the /requests endpoint we fixed):
+         - Login as customer, get token
+         - GET /api/customers/requests — should return HTTP 200 with {"data":[]}
+         - POST a lead request as the photographer, then check GET /api/customers/requests again
+
+      3. CUSTOMER ACCEPT/REJECT LEAD REQUEST:
+         - First, check what the approve/reject endpoints look like
+         - Try POST /api/customers/requests/:lead_request_id/approve
+         - Try POST /api/customers/requests/:lead_request_id/reject
+         - Note: these may fail with 404 or 500 — document what happens
+
+      4. PROFESSIONAL LEAD REQUEST FLOW:
+         - Login as photographer, get fresh token
+         - GET /api/photographers/marketplace — should show OPEN leads
+         - POST /api/photographers/leads/request — send a new lead request
+         - GET /api/photographers/leads/requests/me — list my sent requests
+         - DELETE /api/photographers/leads/requests/:id — cancel a pending request
+
+      Write results to /Users/ashwin/workspace/e2e-results/track-2-leads.md with:
+      - Each endpoint tested, HTTP status, response body
+      - Any errors, 4xx/5xx codes, and what they mean
+      - Overall status of leads + lead request flow
+    assigned_to: general
+    verify_skip_reason: "Downstream compilation task reads all result files directly; no independent verification needed for raw API test output."
+    max_retries: 1
+
+  - id: track-3-payments
+    title: Test payments and tracecoin flows
+    prompt: |
+      You are testing the Nxtgauge backend API. All tests are via direct HTTP API calls.
+
+      Gateway: http://localhost:9100
+      Payments service: http://localhost:9116
+      Test accounts: bug2_test_1781086381@test.com / Test1234!
+
+      DATABASE: postgresql://nxtgauge:nxtgauge_dev@localhost:5432/nxtgauge_db
+
+      Test these flows:
+
+      1. TRACEOCOIN WALLET:
+         - GET /api/packages — list all pricing packages (via gateway or users service)
+         - GET /api/photographers/wallet/me — check photographer wallet (may fail with 404 if not set up)
+         - Check DB: psql "postgresql://nxtgauge:nxtgauge_dev@localhost:5432/nxtgauge_db" -c "SELECT * FROM tracecoin_wallets LIMIT 5;"
+
+      2. TRACEOCOIN PURCHASE (if payments service is running on 9116):
+         - GET http://localhost:9116/health — is payments service up?
+         - GET http://localhost:9116/api/packages — list packages
+         - Try to initiate a tracecoin purchase (check the payments service routes)
+
+      3. PAYMENTS SERVICE ROUTES:
+         - Document what routes the payments service exposes
+         - Try GET /api/payments/health
+         - Try GET /api/payments/packages
+         - Check what the Beeceptor mock URL is (if used)
+
+      4. TRACEOCOIN LEDGER:
+         - Check DB: psql "postgresql://nxtgauge:nxtgauge_dev@localhost:5432/nxtgauge_db" -c "SELECT * FROM tracecoin_ledger LIMIT 10;"
+         - Check the schema of tracecoin_ledger table
+
+      5. INVOICES:
+         - Check DB: psql ... -c "SELECT * FROM invoices LIMIT 5;"
+         - Check invoices table schema
+
+      Write results to /Users/ashwin/workspace/e2e-results/track-3-payments.md with:
+      - Payments service status (running or not)
+      - All endpoints tested with HTTP status and response
+      - Tracecoin wallet balances found in DB
+      - Any payment flow errors
+    assigned_to: general
+    verify_skip_reason: "Downstream compilation task reads all result files directly; no independent verification needed for raw API test output."
+    max_retries: 1
+
+  - id: track-4-admin
+    title: Test admin approval and verification workflow
+    prompt: |
+      You are testing the Nxtgauge backend API. All tests are via direct HTTP API calls.
+
+      Gateway: http://localhost:9100
+      Test admin account: cu_1781041378@test.com / Test1234! (or use admin credentials if available)
+
+      NOTE: The admin dashboard is a separate Next.js app. Test the backend API endpoints only.
+
+      Test these flows:
+
+      1. ADMIN VERIFICATIONS ENDPOINT:
+         - GET /api/admin/verifications — list verification cases (requires admin auth)
+         - GET /api/admin/verifications?status=PENDING — filter by status
+         - What does this return? 200? 403? 401?
+
+      2. ADMIN APPROVALS ENDPOINT:
+         - GET /api/admin/approvals — list approval cases
+         - What does it return?
+
+      3. APPROVE A JOB (job posting approval):
+         - First create a job as a company (use co_1781041378@test.com / Test1234! if available, or register new)
+         - Submit the job (status goes to PENDING_APPROVAL)
+         - Try to approve it via admin endpoint: POST /api/admin/jobs/:job_id/approve or similar
+         - Document what admin job approval endpoints exist
+
+      4. APPROVE A REQUIREMENT (customer lead approval):
+         - Create a customer requirement
+         - Submit it (status goes to PENDING_APPROVAL)
+         - Try to approve via admin endpoint
+
+      5. ADMIN APPROVAL FLOW:
+         - Check what routes exist under /api/admin/ for approvals
+         - GET /api/admin/approval-cases if it exists
+         - Try to find and call the actual approval endpoints
+
+      6. VERIFICATION MANAGEMENT:
+         - GET /api/admin/verifications — list all verifications
+         - Try to approve/reject a verification
+
+      Write results to /Users/ashwin/workspace/e2e-results/track-4-admin.md with:
+      - Each admin endpoint tested, HTTP status, response
+      - Whether verification/approval workflow is functional end-to-end
+      - Missing or broken admin endpoints
+    assigned_to: general
+    verify_skip_reason: "Downstream compilation task reads all result files directly; no independent verification needed for raw API test output."
+    max_retries: 1
+
+  - id: track-5-support
+    title: Test support tickets and email notifications
+    prompt: |
+      You are testing the Nxtgauge backend API. All tests are via direct HTTP API calls.
+
+      Gateway: http://localhost:9100
+      Test accounts: bug2_test_1781086381@test.com / Test1234! (or register fresh users with test_mode=true)
+
+      Test these flows:
+
+      1. SUPPORT TICKETS (user-facing):
+         - POST /api/support/tickets — create a support ticket
+           Body: {"subject":"Test ticket from API","message":"This is a test","category":"GENERAL"}
+         - GET /api/support/tickets — list my tickets
+         - GET /api/support/tickets/:id — get specific ticket
+         - POST /api/support/tickets/:id/messages — add a message to a ticket
+
+      2. SUPPORT TICKETS (admin):
+         - GET /api/admin/support-cases — list all support cases (admin only)
+         - Try to reply to a ticket as admin
+
+      3. EMAIL NOTIFICATIONS (check if emails are being sent):
+         - Create a ticket → check DB: psql "postgresql://nxtgauge:nxtgauge_dev@localhost:5432/nxtgauge_db" -c "SELECT * FROM email_logs ORDER BY created_at DESC LIMIT 10;"
+         - Submit a job → check email_logs
+         - Submit a requirement → check email_logs
+         - Submit a job application → check email_logs
+
+      4. NOTIFICATIONS:
+         - GET /api/me/notifications — list user notifications
+         - Check notification table: psql "postgresql://nxtgauge:nxtgauge_dev@localhost:5432/nxtgauge_db" -c "SELECT * FROM notifications ORDER BY created_at DESC LIMIT 20;"
+
+      5. KNOWLEDGE BASE:
+         - GET /api/kb/articles — list KB articles
+         - GET /api/kb/articles/:id — get specific article
+         - GET /api/kb/categories — list categories
+
+      Write results to /Users/ashwin/workspace/e2e-results/track-5-support.md with:
+      - Support ticket creation status (success/fail/error)
+      - Email logs found in DB (were emails sent?)
+      - Notifications generated
+      - KB articles accessible
+      - Any broken endpoints
+    assigned_to: general
+    verify_skip_reason: "Downstream compilation task reads all result files directly; no independent verification needed for raw API test output."
+    max_retries: 1
+
+  - id: track-6-reviews-portfolio
+    title: Test reviews, portfolio, and onboarding flows
+    prompt: |
+      You are testing the Nxtgauge backend API. All tests are via direct HTTP API calls.
+
+      Gateway: http://localhost:9100
+      Test accounts: register fresh users with test_mode=true as needed
+
+      DATABASE: postgresql://nxtgauge:nxtgauge_dev@localhost:5432/nxtgauge_db
+
+      Test these flows:
+
+      1. REVIEWS:
+         - POST /api/reviews — submit a review
+           Body: {"entity_type":"COMPANY","entity_id":"<company_uuid>","rating":5,"comment":"Great service!"}
+           (You'll need to find a valid company_id from the DB: psql ... -c "SELECT id FROM company_profiles LIMIT 3;")
+         - GET /api/reviews — list reviews
+         - GET /api/reviews?entity_type=COMPANY&entity_id=<id> — reviews for a company
+         - Admin: GET /api/admin/reviews — admin review list
+
+      2. PORTFOLIO (photographer):
+         - Login as photographer (bug2_test_1781086381@test.com / Test1234!)
+         - POST /api/photographers/portfolio/me — create portfolio item
+           Body: {"title":"Test Portfolio","description":"Test description","image_url":"https://example.com/test.jpg"}
+         - GET /api/photographers/portfolio/me — list portfolio items
+         - PATCH /api/photographers/portfolio/me/:id — update portfolio item
+         - DELETE /api/photographers/portfolio/me/:id — delete portfolio item
+
+      3. SERVICES (photographer):
+         - POST /api/photographers/services/me — create a service
+           Body: {"name":"Wedding Photography Package","description":"Full day coverage","price_inr":50000}
+         - GET /api/photographers/services/me — list services
+         - PATCH /api/photographers/services/me/:id — update service
+         - DELETE /api/photographers/services/me/:id — delete service
+
+      4. ONBOARDING STATE:
+         - GET /api/onboarding/state — get onboarding state for a user
+         - POST /api/onboarding/submit — submit onboarding data
+         - GET /api/me — check if role and profile status is shown
+
+      5. RUNTIME CONFIG:
+         - GET /api/runtime-config — get runtime config for a role
+         - GET /api/runtime-config?role=PHOTOGRAPHER — get specific role config
+
+      Write results to /Users/ashwin/workspace/e2e-results/track-6-reviews-portfolio.md with:
+      - Each endpoint tested, HTTP status, response
+      - DB schema for reviews and portfolio tables
+      - Any errors found
+    assigned_to: general
+    verify_skip_reason: "Downstream compilation task reads all result files directly; no independent verification needed for raw API test output."
+    max_retries: 1
+
+  - id: track-7-compile-report
+    title: Compile comprehensive test report
+    role: verify-as-task
+    prompt: |
+      Read all the test result files created by the other tracks and compile a comprehensive report.
+
+      Read these files:
+      - /Users/ashwin/workspace/e2e-results/track-1-professions.md
+      - /Users/ashwin/workspace/e2e-results/track-2-leads.md
+      - /Users/ashwin/workspace/e2e-results/track-3-payments.md
+      - /Users/ashwin/workspace/e2e-results/track-4-admin.md
+      - /Users/ashwin/workspace/e2e-results/track-5-support.md
+      - /Users/ashwin/workspace/e2e-results/track-6-reviews-portfolio.md
+      - Also read: /Users/ashwin/workspace/E2E_TEST_REPORT.md (the previous report from Session 1)
+
+      Compile everything into a single comprehensive report at:
+      /Users/ashwin/workspace/e2e-results/COMPLETE_E2E_REPORT.md
+
+      The report should include:
+      1. Executive summary: overall system health (X/Y endpoints working)
+      2. Track-by-track results with PASS/FAIL for each endpoint tested
+      3. All bugs found this round (new issues not in the previous report)
+      4. All bugs from the previous report and their current status (fixed/still broken)
+      5. Recommendations and priority list of what to fix next
+      6. Test data created during this round
+
+      Format: Markdown report. Be thorough — this is the canonical test record.
+    assigned_to: general
+    depends_on:
+      - track-1-professions
+      - track-2-leads
+      - track-3-payments
+      - track-4-admin
+      - track-5-support
+      - track-6-reviews-portfolio
+    max_retries: 1

Cargo.lock

@@ -941,6 +941,7 @@ dependencies = [
  "chrono",
  "db",
  "email",
+ "reqwest",
  "serde",
  "serde_json",
  "sqlx",

apps/cron/Cargo.toml

@@ -18,3 +18,4 @@ serde.workspace = true
 serde_json.workspace = true
 chrono.workspace = true
 uuid.workspace = true
+reqwest.workspace = true

apps/cron/src/main.rs

@@ -75,6 +75,19 @@ async fn main() -> Result<(), Box<dyn std::error::Error>> {
         }
     });
 
+    // Spawn Hourly Auto-Apply task for job seekers
+    let p_auto_apply_sys = pool.clone();
+    tokio::spawn(async move {
+        let mut interval = time::interval(Duration::from_secs(60 * 60));
+        loop {
+            interval.tick().await;
+            tracing::info!("Running Auto-Apply Task...");
+            if let Err(e) = tasks::auto_apply::run_auto_apply(&p_auto_apply_sys).await {
+                tracing::error!("Auto-Apply Task Failed: {}", e);
+            }
+        }
+    });
+
     // Keep main thread alive
     tokio::signal::ctrl_c().await?;
     tracing::info!("Shutting down cron engine.");

apps/cron/src/tasks/auto_apply.rs

@@ -0,0 +1,318 @@
+use chrono::{Duration, Utc};
+use reqwest::Client;
+use serde::{Deserialize, Serialize};
+use sqlx::PgPool;
+use uuid::Uuid;
+
+#[derive(Debug, Deserialize)]
+struct AutoApplyConfig {
+    litellm_base_url: String,
+    litellm_api_key: String,
+    litellm_model: String,
+    max_applications_per_run: usize,
+    max_daily_applications: usize,
+}
+
+impl AutoApplyConfig {
+    fn from_env() -> Self {
+        Self {
+            litellm_base_url: std::env::var("LITELLM_BASE_URL")
+                .unwrap_or_else(|_| "https://llm.nxtgauge.com/v1".to_string()),
+            litellm_api_key: std::env::var("LITELLM_API_KEY").unwrap_or_default(),
+            litellm_model: std::env::var("LITELLM_MODEL")
+                .unwrap_or_else(|_| "askash-main".to_string()),
+            max_applications_per_run: std::env::var("AUTO_APPLY_MAX_PER_RUN")
+                .unwrap_or_else(|_| "5".to_string())
+                .parse()
+                .unwrap_or(5),
+            max_daily_applications: std::env::var("AUTO_APPLY_MAX_DAILY")
+                .unwrap_or_else(|_| "20".to_string())
+                .parse()
+                .unwrap_or(20),
+        }
+    }
+}
+
+#[derive(Debug, Serialize)]
+struct ChatMessage {
+    role: String,
+    content: String,
+}
+
+#[derive(Debug, Serialize)]
+struct ChatCompletionRequest {
+    model: String,
+    messages: Vec<ChatMessage>,
+    temperature: f32,
+    max_tokens: i32,
+}
+
+#[derive(Debug, Deserialize)]
+struct ChatCompletionResponse {
+    choices: Vec<Choice>,
+}
+
+#[derive(Debug, Deserialize)]
+struct Choice {
+    message: Message,
+}
+
+#[derive(Debug, Deserialize)]
+struct Message {
+    content: String,
+}
+
+async fn generate_cover_letter(
+    client: &Client,
+    config: &AutoApplyConfig,
+    seeker_name: &str,
+    experience: i32,
+    skills: &[String],
+    summary: Option<&str>,
+    job_title: &str,
+    job_desc: &str,
+) -> Result<String, Box<dyn std::error::Error + Send + Sync>> {
+    let url = format!("{}/chat/completions", config.litellm_base_url.trim_end_matches('/'));
+
+    let prompt = format!(
+        "Write a brief, professional cover letter (max 200 words).\n\n\
+         IMPORTANT: Do NOT include phone number, email, or any contact information.\n\
+         Only use the information provided below.\n\n\
+         CANDIDATE: Name: {}, Experience: {} years, Skills: {}, Summary: {}\n\
+         JOB: Title: {}, Description: {}\n\n\
+         Cover Letter:",
+        seeker_name,
+        experience,
+        skills.join(", "),
+        summary.unwrap_or(""),
+        job_title,
+        job_desc
+    );
+
+    let payload = ChatCompletionRequest {
+        model: config.litellm_model.clone(),
+        messages: vec![
+            ChatMessage {
+                role: "system".to_string(),
+                content: "You are a professional cover letter writer.".to_string(),
+            },
+            ChatMessage {
+                role: "user".to_string(),
+                content: prompt,
+            },
+        ],
+        temperature: 0.2,
+        max_tokens: 500,
+    };
+
+    let res = client
+        .post(&url)
+        .header("Authorization", format!("Bearer {}", config.litellm_api_key))
+        .header("Content-Type", "application/json")
+        .json(&payload)
+        .send()
+        .await?;
+
+    if !res.status().is_success() {
+        return Ok("I am excited to apply for this position.".to_string());
+    }
+
+    let body: ChatCompletionResponse = res.json().await?;
+
+    if let Some(choice) = body.choices.first() {
+        Ok(choice.message.content.trim().to_string())
+    } else {
+        Ok("I am excited to apply for this position.".to_string())
+    }
+}
+
+#[derive(Debug, sqlx::FromRow)]
+struct JobSeekerWithAi {
+    user_id: Uuid,
+    profile_id: Uuid,
+    full_name: String,
+    skills: Vec<String>,
+    experience_years: i32,
+    summary: Option<String>,
+    daily_limit: i32,
+    used_today: i32,
+}
+
+#[derive(Debug, sqlx::FromRow)]
+struct NewJob {
+    id: Uuid,
+    title: String,
+    description: String,
+    skills: Vec<String>,
+}
+
+pub async fn run_auto_apply(pool: &PgPool) -> Result<(), Box<dyn std::error::Error + Send + Sync>> {
+    let config = AutoApplyConfig::from_env();
+
+    if config.litellm_api_key.is_empty() {
+        tracing::warn!("Auto-apply skipped: LITELLM_API_KEY not configured");
+        return Ok(());
+    }
+
+    tracing::info!("Starting auto-apply job...");
+
+    let client = Client::new();
+    let cutoff_time = Utc::now() - Duration::hours(24);
+
+    let seekers: Vec<JobSeekerWithAi> = sqlx::query_as(
+        r#"
+        SELECT
+            u.id as user_id,
+            jsp.id as profile_id,
+            COALESCE(CONCAT(jsp.first_name, ' ', jsp.last_name), 'Candidate') as full_name,
+            COALESCE(jsp.skills, ARRAY[]::text[]) as skills,
+            COALESCE(jsp.experience_years, 0) as experience_years,
+            jsp.summary,
+            COALESCE((
+                SELECT p.monthly_action_limit / 30
+                FROM ai_entitlements e
+                JOIN ai_plans p ON p.id = e.plan_id
+                WHERE e.user_id = u.id AND e.role = 'JOB_SEEKER' AND e.status = 'active'
+                ORDER BY e.valid_from DESC
+                LIMIT 1
+            ), 10) as daily_limit,
+            COALESCE((
+                SELECT generations_used
+                FROM job_seeker_ai_usage jau
+                WHERE jau.job_seeker_id = jsp.id
+                AND usage_date = CURRENT_DATE
+            ), 0) as used_today
+        FROM users u
+        INNER JOIN job_seeker_profiles jsp ON jsp.user_id = u.id
+        WHERE u.status = 'ACTIVE'
+        AND jsp.status IN ('APPROVED', 'PENDING_REVIEW')
+        AND jsp.skills IS NOT NULL
+        AND array_length(jsp.skills, 1) > 0
+        AND u.email_verified = true
+        "#
+    )
+    .fetch_all(pool)
+    .await?;
+
+    if seekers.is_empty() {
+        tracing::info!("No job seekers with AI packs found for auto-apply");
+        return Ok(());
+    }
+
+    tracing::info!("Found {} job seekers eligible for auto-apply", seekers.len());
+
+    let mut total_applications = 0;
+
+    for seeker in seekers {
+        let remaining_today = seeker.daily_limit - seeker.used_today;
+        if remaining_today <= 0 {
+            tracing::debug!("User {} has no remaining AI applications today", seeker.user_id);
+            continue;
+        }
+
+        let new_jobs: Vec<NewJob> = sqlx::query_as(
+            r#"
+            SELECT j.id, j.title, j.description, COALESCE(j.skills, ARRAY[]::text[]) as skills
+            FROM jobs j
+            INNER JOIN companies c ON c.id = j.company_id
+            WHERE j.status = 'LIVE'
+            AND j.created_at > $1
+            AND c.status = 'APPROVED'
+            AND j.id NOT IN (
+                SELECT job_id FROM job_applications WHERE applicant_user_id = $2
+            )
+            AND j.skills && $3::text[]
+            ORDER BY j.created_at DESC
+            LIMIT $4
+            "#
+        )
+        .bind(cutoff_time)
+        .bind(seeker.user_id)
+        .bind(&seeker.skills)
+        .bind(std::cmp::min(remaining_today as i32, config.max_applications_per_run as i32))
+        .fetch_all(pool)
+        .await?;
+
+        if new_jobs.is_empty() {
+            continue;
+        }
+
+        tracing::info!(
+            "User {} has {} matching new jobs, applying to {}",
+            seeker.user_id,
+            new_jobs.len(),
+            new_jobs.len().min(remaining_today as usize)
+        );
+
+        for job in new_jobs.into_iter().take(remaining_today as usize) {
+            let cover_letter = match generate_cover_letter(
+                &client,
+                &config,
+                &seeker.full_name,
+                seeker.experience_years,
+                &seeker.skills,
+                seeker.summary.as_deref(),
+                &job.title,
+                &job.description,
+            )
+            .await
+            {
+                Ok(cl) => cl,
+                Err(e) => {
+                    tracing::warn!("Failed to generate cover letter: {}", e);
+                    "I am excited to apply for this position.".to_string()
+                }
+            };
+
+            let result = sqlx::query(
+                r#"
+                INSERT INTO job_applications (job_id, applicant_user_id, cover_letter, applied_via_ai)
+                VALUES ($1, $2, $3, true)
+                ON CONFLICT (job_id, applicant_user_id) DO NOTHING
+                "#
+            )
+            .bind(job.id)
+            .bind(seeker.user_id)
+            .bind(&cover_letter)
+            .execute(pool)
+            .await;
+
+            match result {
+                Ok(r) => {
+                    if r.rows_affected() > 0 {
+                        total_applications += 1;
+
+                        sqlx::query(
+                            r#"
+                            INSERT INTO job_seeker_ai_usage (job_seeker_id, usage_date, generations_used)
+                            VALUES ($1, CURRENT_DATE, 1)
+                            ON CONFLICT (job_seeker_id, usage_date)
+                            DO UPDATE SET generations_used = job_seeker_ai_usage.generations_used + 1
+                            "#
+                        )
+                        .bind(seeker.profile_id)
+                        .execute(pool)
+                        .await
+                        .ok();
+
+                        sqlx::query(
+                            "UPDATE job_seeker_profiles SET active_application_count = active_application_count + 1 WHERE id = $1"
+                        )
+                        .bind(seeker.profile_id)
+                        .execute(pool)
+                        .await
+                        .ok();
+
+                        tracing::info!("Auto-applied user {} to job {}", seeker.user_id, job.id);
+                    }
+                }
+                Err(e) => {
+                    tracing::error!("Failed to insert application: {}", e);
+                }
+            }
+        }
+    }
+
+    tracing::info!("Auto-apply completed. Total applications: {}", total_applications);
+    Ok(())
+}

apps/cron/src/tasks/mod.rs

@@ -2,3 +2,4 @@ pub mod leads;
 pub mod requirements;
 pub mod jobs;
 pub mod reminders;
+pub mod auto_apply;

apps/users/src/handlers/ai.rs

@@ -51,32 +51,7 @@ struct OllamaGenerateResponse {
 
 async fn call_ollama(_state: &AppState, model: &str, prompt: &str) -> Result<String, String> {
     let base_url = std::env::var("OLLAMA_BASE_URL").unwrap_or_else(|_| "http://ollama.nxtgauge-ai.svc.cluster.local:11434".to_string());
-    let url = format!("{}/api/generate", base_url);
-
-    let req = OllamaGenerateRequest {
-        model: model.to_string(),
-        prompt: prompt.to_string(),
-        stream: false,
-    };
-
-    let client = reqwest::Client::new();
-    let response = client
-        .post(&url)
-        .json(&req)
-        .send()
-        .await
-        .map_err(|e| format!("ollama request failed: {}", e))?;
-
-    if !response.status().is_success() {
-        return Err(format!("ollama returned status: {}", response.status()));
-    }
-
-    let result: OllamaGenerateResponse = response
-        .json()
-        .await
-        .map_err(|e| format!("failed to parse ollama response: {}", e))?;
-
-    Ok(result.response)
+    call_ollama_inline(&base_url, model, prompt).await
 }
 
 // ── Phase 1: Strict keyword fast-path for intent classification ────────────────
@@ -318,7 +293,39 @@ fn is_internal_admin(auth: &AuthUser) -> bool {
         || auth.claims.roles.contains(&"SUPER_ADMIN".to_string())
 }
 
-async fn call_ollama_inline(base_url: &str, model: &str, prompt: &str) -> Result<String, String> {
+fn get_llm_base_url() -> String {
+    let llm_provider = std::env::var("LLM_PROVIDER").unwrap_or_else(|_| "ollama".to_string());
+    if llm_provider == "litellm" {
+        std::env::var("LITELLM_BASE_URL")
+            .unwrap_or_else(|_| "https://llm.nxtgauge.com/v1".to_string())
+    } else {
+        std::env::var("OLLAMA_BASE_URL")
+            .unwrap_or_else(|_| "http://ollama.nxtgauge-ai.svc.cluster.local:11434".to_string())
+    }
+}
+
+fn get_llm_model() -> String {
+    let llm_provider = std::env::var("LLM_PROVIDER").unwrap_or_else(|_| "ollama".to_string());
+    if llm_provider == "litellm" {
+        std::env::var("LITELLM_MODEL")
+            .unwrap_or_else(|_| "askash-main".to_string())
+    } else {
+        std::env::var("OLLAMA_CHAT_MODEL")
+            .unwrap_or_else(|_| "gemma3:270m".to_string())
+    }
+}
+
+async fn call_ollama_inline(_base_url: &str, _model: &str, prompt: &str) -> Result<String, String> {
+    let llm_provider = std::env::var("LLM_PROVIDER").unwrap_or_else(|_| "ollama".to_string());
+
+    if llm_provider == "litellm" {
+        let base_url = get_llm_base_url();
+        let model = get_llm_model();
+        return call_litellm_inline(&base_url, &model, prompt).await;
+    }
+
+    let base_url = get_llm_base_url();
+    let model = get_llm_model();
     let url = format!("{}/api/generate", base_url);
     let req = OllamaGenerateRequest {
         model: model.to_string(),
@@ -346,6 +353,77 @@ async fn call_ollama_inline(base_url: &str, model: &str, prompt: &str) -> Result
     Ok(result.response)
 }
 
+#[derive(Debug, Serialize)]
+struct LiteLLMChatMessage {
+    role: String,
+    content: String,
+}
+
+#[derive(Debug, Serialize)]
+struct LiteLLMRequest {
+    model: String,
+    messages: Vec<LiteLLMChatMessage>,
+    temperature: f32,
+    max_tokens: i32,
+}
+
+#[derive(Debug, Deserialize)]
+struct LiteLLMResponse {
+    choices: Vec<LiteLLMChoice>,
+}
+
+#[derive(Debug, Deserialize)]
+struct LiteLLMChoice {
+    message: LiteLLMMessage,
+}
+
+#[derive(Debug, Deserialize)]
+struct LiteLLMMessage {
+    content: String,
+}
+
+async fn call_litellm_inline(base_url: &str, model: &str, prompt: &str) -> Result<String, String> {
+    let api_key = std::env::var("LITELLM_API_KEY").unwrap_or_default();
+    let url = format!("{}/chat/completions", base_url.trim_end_matches('/'));
+
+    let payload = LiteLLMRequest {
+        model: model.to_string(),
+        messages: vec![
+            LiteLLMChatMessage {
+                role: "user".to_string(),
+                content: prompt.to_string(),
+            },
+        ],
+        temperature: 0.2,
+        max_tokens: 2048,
+    };
+
+    let client = reqwest::Client::new();
+    let response = client
+        .post(&url)
+        .header("Authorization", format!("Bearer {}", api_key))
+        .header("Content-Type", "application/json")
+        .json(&payload)
+        .send()
+        .await
+        .map_err(|e| format!("litellm request failed: {}", e))?;
+
+    if !response.status().is_success() {
+        return Err(format!("litellm returned status: {}", response.status()));
+    }
+
+    let result: LiteLLMResponse = response
+        .json()
+        .await
+        .map_err(|e| format!("failed to parse litellm response: {}", e))?;
+
+    if let Some(choice) = result.choices.first() {
+        Ok(choice.message.content.trim().to_string())
+    } else {
+        Err("litellm returned empty response".to_string())
+    }
+}
+
 async fn ai_chat_message(
     State(state): State<AppState>,
     Json(body): Json<OllamaChatRequest>,
@@ -355,13 +433,13 @@ async fn ai_chat_message(
         return (status, Json(payload)).into_response();
     }
 
-    let ollama_base = std::env::var("OLLAMA_BASE_URL").unwrap_or_else(|_| "http://ollama.nxtgauge-ai.svc.cluster.local:11434".to_string());
-    let model = std::env::var("OLLAMA_CHAT_MODEL").unwrap_or_else(|_| "gemma3:270m".to_string());
+    let ollama_base = get_llm_base_url();
+    let model = get_llm_model();
     let default_conversation = Uuid::new_v4().to_string();
 
     let conversation_id = body.conversation_id.unwrap_or_else(|| default_conversation);
 
-    // ── Phase 1: Strict keyword fast-path (skips Ollama when unambiguous) ─────
+    // ── Phase 1: Strict keyword fast-path (skips LLM when unambiguous) ─────
     let (intent, confidence) = match classify_strict_keywords(&body.message) {
         Some((kw_intent, kw_conf)) => (kw_intent.to_string(), kw_conf),
         None => classify_intent(&body.message, &ollama_base, &model).await,
@@ -606,8 +684,8 @@ async fn ai_extract_form(
     State(_state): State<AppState>,
     Json(body): Json<FormExtractBody>,
 ) -> impl IntoResponse {
-    let ollama_base = std::env::var("OLLAMA_BASE_URL").unwrap_or_else(|_| "http://ollama.nxtgauge-ai.svc.cluster.local:11434".to_string());
-    let model = std::env::var("OLLAMA_CHAT_MODEL").unwrap_or_else(|_| "gemma3:270m".to_string());
+    let ollama_base = get_llm_base_url();
+    let model = get_llm_model();
 
     let form_type = body.form_type.unwrap_or_else(|| "generic".to_string());
 
@@ -834,8 +912,8 @@ async fn ai_generate_job_field(
         }
     };
 
-    let ollama_base = std::env::var("OLLAMA_BASE_URL").unwrap_or_else(|_| "http://ollama.nxtgauge-ai.svc.cluster.local:11434".to_string());
-    let model = std::env::var("OLLAMA_CHAT_MODEL").unwrap_or_else(|_| "gemma3:270m".to_string());
+    let ollama_base = get_llm_base_url();
+    let model = get_llm_model();
 
     let field_prompt = match body.field.as_str() {
         "title" => format!(
@@ -949,8 +1027,8 @@ async fn ai_generate_cover_letter(
         }
     };
 
-    let ollama_base = std::env::var("OLLAMA_BASE_URL").unwrap_or_else(|_| "http://ollama.nxtgauge-ai.svc.cluster.local:11434".to_string());
-    let model = std::env::var("OLLAMA_CHAT_MODEL").unwrap_or_else(|_| "gemma3:270m".to_string());
+    let ollama_base = get_llm_base_url();
+    let model = get_llm_model();
 
     let notes = body.additional_notes.as_deref().unwrap_or("");
     let skills_str = skills.join(", ");
@@ -1062,8 +1140,8 @@ async fn ai_tailor_resume(
         }
     };
 
-    let ollama_base = std::env::var("OLLAMA_BASE_URL").unwrap_or_else(|_| "http://ollama.nxtgauge-ai.svc.cluster.local:11434".to_string());
-    let model = std::env::var("OLLAMA_CHAT_MODEL").unwrap_or_else(|_| "gemma3:270m".to_string());
+    let ollama_base = get_llm_base_url();
+    let model = get_llm_model();
 
     let existing_resume = body.resume_text.as_deref().unwrap_or("Not provided");
     let skills_str = skills.join(", ");
@@ -1186,8 +1264,8 @@ async fn ai_auto_apply(
         return (StatusCode::TOO_MANY_REQUESTS, Json(serde_json::json!({ "error": format!("Only {} generations left today", remaining) }))).into_response();
     }
 
-    let ollama_base = std::env::var("OLLAMA_BASE_URL").unwrap_or_else(|_| "http://ollama.nxtgauge-ai.svc.cluster.local:11434".to_string());
-    let model = std::env::var("OLLAMA_CHAT_MODEL").unwrap_or_else(|_| "gemma3:270m".to_string());
+    let ollama_base = get_llm_base_url();
+    let model = get_llm_model();
     let skills_str = skills.join(", ");
 
     let mut created = 0;