Fraz B.

Voice AI Agents | RAG Systems | AI Chatbot | AI SaaS

𝗬𝗼𝘂 𝗻𝗲𝗲𝗱 𝗮𝗻 𝗔𝗜 𝘀𝘆𝘀𝘁𝗲𝗺 𝘆𝗼𝘂𝗿 𝘁𝗲𝗮𝗺 𝗰𝗮𝗻 𝘁𝗿𝘂𝘀𝘁 𝘄𝗶𝘁𝗵 𝗹𝗶𝘃𝗲 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿𝘀 𝘁𝗼𝗺𝗼𝗿𝗿𝗼𝘄 - 𝗻𝗼𝘁 𝗮 𝗱𝗲𝗺𝗼 𝘁𝗵𝗮𝘁 𝗯𝗿𝗲𝗮𝗸𝘀 𝘁𝗵𝗲 𝗺𝗼𝗺𝗲𝗻𝘁 𝗿𝗲𝗮𝗹 𝘂𝘀𝗲𝗿𝘀 𝘁𝗼𝘂𝗰𝗵 𝗶𝘁. I built my first Voice AI agent 24 months ago. It worked perfectly in testing. Then a caller with a heavy accent asked a question while their dog barked in the background, and the whole thing collapsed. That project taught me something most AI developers learn too late: the demo is 10% of the work. The other 90% is everything that can go wrong - bad audio, CRM timeouts, users who say "um" fourteen times, edge cases nobody warned you about. Now I only build AI that handles the chaos. 𝗪𝗵𝗮𝘁 𝗜'𝘃𝗲 𝗯𝘂𝗶𝗹𝘁 𝗩𝗮𝗿𝗽𝗮 is a HIPAA-compliant healthcare voice agent for multi-clinic practices. It answers calls 24/7, classifies caller intent across six categories, routes to the right staff, and books appointments. 2000+ calls handled in six weeks at a 100% pickup rate. Built HIPAA-aware with no PHI in logs and EMR integration. 𝗔𝗲𝗹𝗼 is a productized AI voice receptionist for dental practices, live 24/7 with paying customers across three subscription tiers. It answers every call, books and reschedules appointments, handles FAQs, and escalates to staff only when needed, capturing the after-hours patients clinics were losing to voicemail. 𝗩𝗲𝗿𝗱𝗮𝗽𝘁 is a multi-tenant voice AI platform for home-service businesses like landscaping, HVAC, and plumbing. Each branded agent answers calls 24/7, books and qualifies jobs, syncs to Google and Outlook calendars, logs everything in GoHighLevel, and fires invoices automatically once the owner approves a job. 𝗤𝗮𝗹𝗶𝗳𝗔𝗜 takes 40+ real estate calls daily. Not test calls - actual leads, live conversations, 92% intent accuracy measured across thousands of interactions. Built on VAPI with ElevenLabs voice, GoHighLevel CRM sync, and Google Calendar booking. It qualifies buyers, handles objections, and hands off to human agents without dropping context. 𝗣𝗮𝗽𝘆𝗿𝘂𝘀 answers research questions across 15,000+ academic papers. Hierarchical chunking, hybrid search with pgvector, Cohere reranking, and cited answers. It says "I don't know" when retrieval confidence is low. Researchers trust it because it doesn't make things up. 𝗔𝗜 𝗙𝗿𝗮𝗻𝗸 generates legal documents with jurisdiction-aware retrieval. 700+ documents created with proper citations pulled from the right statutes. React frontend, Python backend, pgvector for legal precedent matching. 𝗦𝘁𝗮𝘆𝗦𝗾𝘂𝗶𝗿𝗲 handles hotel reception calls - booking inquiries, room availability, guest requests. Retell for voice, n8n workflows for automation, webhook integrations with property management systems. 𝗛𝗼𝘄 𝗜 𝘄𝗼𝗿𝗸 I take on Voice AI agents, RAG systems, and AI MVPs. I'm useful when you're past the "exploring AI" phase and need something deployed. Less useful if you're still figuring out whether AI makes sense - though I'm happy to tell you if it doesn't. Most of my projects are $2K–25K and take 2–6 weeks. I don't disappear mid-project, I document everything, and I build with handoff in mind so you're not dependent on me forever. 𝗧𝗲𝗰𝗵 𝗜 𝘄𝗼𝗿𝗸 𝘄𝗶𝘁𝗵 𝗩𝗼𝗶𝗰𝗲 𝗔𝗜: VAPI, Retell, Twilio, Bland, ElevenLabs, OpenAI Whisper, OpenAI TTS, Deepgram 𝗥𝗔𝗚 & 𝗟𝗟𝗠𝘀: LangChain, LangGraph, LlamaIndex, OpenAI, Claude, Gemini, Cohere, pgvector, Qdrant, Pinecone, Elasticsearch 𝗕𝗮𝗰𝗸𝗲𝗻𝗱: Python, FastAPI, Node.js, TypeScript, PostgreSQL, Redis, Supabase 𝗙𝗿𝗼𝗻𝘁𝗲𝗻𝗱: React, Next.js, Tailwind CSS 𝗔𝘂𝘁𝗼𝗺𝗮𝘁𝗶𝗼𝗻 & 𝗜𝗻𝘁𝗲𝗴𝗿𝗮𝘁𝗶𝗼𝗻𝘀: n8n, Make, Zapier, GoHighLevel, HubSpot, Google Calendar, Stripe, WhatsApp Business API 𝗗𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁: AWS, GCP, Vercel, Railway, Docker 𝗪𝗵𝗮𝘁 𝗰𝗹𝗶𝗲𝗻𝘁𝘀 𝘀𝗮𝘆 "Fraz is excellent to work with. Fast, thorough, and quick to accept feedback. His team is great, and we have loved working with them to create multiple software solutions. Fantastic Upwork contractor." — Matt Gray, Founder @ Hazard House If any of this sounds relevant, message me with what you're building. I respond within 30 minutes and I'll tell you honestly whether I can help.