AI Engine

Gemini 2.0 Flash — Discovery, intelligence, verification

Model: Gemini 2.0 Flash via @google/genai SDK · Server-side only · Temperature: 0.3 for factual accuracy

Hostel Discovery Engine

POST /api/hostels/search

The discovery engine combines Gemini 2.0 Flash with Google Search Retrieval to find real, verified hostels matching a nomad's skills and desired location. Every result undergoes a multi-stage verification pipeline before being presented to the user.


  User Input                   AI Layer                    Verification Layer
  ──────────                   ────────                    ──────────────────
                                                        
  Location: "Bali"     →   Gemini 2.0 Flash      →    verifyWebsite()
  Skills: "Marketing"      + Google Search              HTTP HEAD each URL
                           Retrieval Tool               HTTPS → HTTP fallback
                                    │                          │
                                    ▼                          ▼
                           15-20 hostel results →    scrapeContactEmails()
                           names, websites,          Crawl /, /contact, /about
                           locations, match %        Regex + mailto: extraction
                                                           │
                                                           ▼
                                                    verifyEmailDomain()
                                                    DNS MX record lookup
                                                    per email address
                                                           │
                                                           ▼
                                                    ✅ Verified Results
                                                    Cached in Supabase

Request / Response

Request Body

{
  "location": "Bali, Indonesia",
  "skills": ["marketing", "SEO",
             "content creation"],
  "userId": "user_2x4k..."
}

Response (per hostel)

{
  "name": "The Mad Monkey",
  "location": "Ubud, Bali",
  "website": "https://themadmonkey.com",
  "websiteVerified": true,
  "contactEmail": "info@themadmonkey.com",
  "emailVerified": true,
  "matchScore": 87,
  "needs": ["social media", "SEO"]
}

Dossier Generation

POST /api/hostels/dossier

Generates a comprehensive market intelligence report for a specific hostel, personalized to the nomad's profile. Includes competitive analysis, needs assessment, strategic positioning, and an auto-generated pitch email ready to send.

Input Parameters

Field	Type	Description
hostelName	string	Name of the hostel
hostelWebsite	string	Primary website URL
userName	string	Nomad's full name
userHeadline	string	Professional headline
userSkills	string[]	Skill list
userBio	string	Professional biography
userLocation	string	Current location

Output Sections

Market Snapshot

Competitive positioning, local market analysis, hostel's digital presence assessment

Needs Analysis

Identified hostel needs that match the nomad's specific skillset

Value Proposition

Why this specific nomad is uniquely positioned for this hostel

Deal Structure

Suggested duration, deliverables, accommodation type

Personalized Pitch

Auto-generated email ready to send — personalized to both parties

Verified Contacts

Scraped + MX-verified email addresses from the hostel's website

Email Verification Pipeline

lib/email-verify.ts

verifyWebsite(url)

HTTP HEAD request to validate hostel URL. Tries HTTPS first, falls back to HTTP. Returns final resolved URL or null.

scrapeContactEmails(url)

Fetches the hostel website's /, /contact, and /about pages. Extracts emails via regex patterns and mailto: link parsing.

verifyEmailDomain(email)

DNS MX record lookup via Node.js dns.resolveMx(). Confirms the email domain has valid mail exchange servers.

verifyAndEnrichHostels(hostels)

Orchestration function: verify website → scrape emails → MX-verify all emails. Returns only verified data.

Anti-Hallucination Measures

LLMs can fabricate URLs and email addresses. HostelHack implements a multi-layer verification pipeline to ensure every piece of data surfaced to users is real.

Layer	Method	What it catches
Prompt Engineering	"ONLY include emails you actually find in search results. Do NOT guess."	Reduces generation of fake data
Website Verification	HTTP HEAD request to every URL	Catches hallucinated/dead URLs
Email Scraping	Real website crawling for contact emails	Replaces AI-guessed emails with real ones
DNS MX Validation	dns.resolveMx() per email domain	Confirms mail servers exist
Placeholder Stripping	Regex replaces "[Your Name]", etc.	Removes template artifacts from pitches
Cache Layer	Search results cached in Supabase	Prevents repeated unverified generations

Result: Users only see hostel URLs that respond to HTTP requests and email addresses whose domains have valid MX records. This dramatically reduces bounce rates and ensures proposals reach real inboxes.