Jameson Sansone

A full-stack RAG application that transforms my grandfather's 200-page WWII memoir into an interactive conversational experience, built for his 103rd birthday.

• RAG Engine: LlamaIndex and FastEmbed for local retrieval; GPT-4o for summarizing the response.
• Custom Voice: ElevenLabs voice cloning of my grandpa's voice for immersive book excerpt readings.
• Architecture: Decoupled FastAPI backend on Render with a responsive Vercel frontend.

• Python
• LlamaIndex
• GPT-4o
• ElevenLabs
• Vercel
• Render

An end-to-end n8n workflow that leverages Apify to mine LinkedIn job listings for growth signals at small plumbing businesses, then generates 1:1 personalized outreach.

• Signal Capture: Apify's LinkedIn Jobs Scraper pulls plumbing companies hiring for Sales, Dispatch, or Operations roles, a proxy for shops outgrowing their manual processes.
• Qualification & Contact Resolution: Rule-based filters narrow to US-based shops under 200 employees, then a GPT-4o-mini classifier buckets each job description as High, Low, or No Intent; qualified leads pass to AnyMailFinder's /decision-maker endpoint to resolve verified Sales contacts via webhook.
• Research-Driven Personalization: Perplexity sonar-pro runs a per-company query prioritizing geographic expansion, public recognition, and team growth, then GPT-4o converts that structured JSON into a single observational opener with cascading fallback logic.

• n8n
• Apify
• AnyMailFinder
• Perplexity Sonar
• OpenAI GPT-4o

A fully self-hosted enrichment proof-of-concept that auto-researches new HubSpot contacts and writes structured intelligence back to the record. Deployed end-to-end on Railway.

• Infrastructure: Railway's "n8n with workers" template provisions a four-service production architecture (n8n Primary, Worker, Postgres, Redis).
• Orchestration: n8n listens for HubSpot contact.creation webhooks, fetches the full record, calls the Python service and writes results back via the native HubSpot node.
• Enrichment: A stateless FastAPI endpoint runs a domain-aware Perplexity sonar-pro query to research the partner's services and territory, returning a normalized summary and fit score.

Next iteration: a human-in-the-loop review queue before CRM write-back to catch hallucinated summaries and edge-case misclassifications before they hit the source of truth.

• Self-hosted n8n on Railway
• Python
• Perplexity Sonar
• HubSpot

A Streamlit tool that shows how Gemini expands a search query, which sources it cites, and how often a target domain appears in AI-generated answers.

• Query Fan-Out: Runs the same query multiple times through Gemini with Google Search grounding to capture hidden subqueries and citation patterns.
• Share of Voice: Labels cited URLs as Target or Other, then calculates how much visibility a chosen domain has across Gemini's cited sources.
• Content Gap Analysis: Compares top-cited competitor pages against the target page using embeddings to surface on-page opportunities.

• Python
• Streamlit
• Gemini API
• Google Search Grounding
• pandas
• sentence-transformers
• scikit-learn

A two-part n8n workflow to handle post-event lead intake. Automatically enrich prospects against ICP criteria leveraging Firecrawl, then send enriched prospects to Google Sheets for a rep to provide a HITL review. The user can choose which prospects to add to HubSpot through an Apps Script integration.

• Intake & Enrichment: Badge-scan data is parsed and enriched against ICP criteria using Firecrawl.
• Human-in-the-Loop: Enriched prospects land in Google Sheets where a rep reviews and selects who advances.
• CRM Sync: Selected prospects are pushed to HubSpot through an Apps Script integration.

• n8n
• Firecrawl
• Google Sheets
• Apps Script
• HubSpot