Kian Kashfipour

Yourang is a B2B AI voice assistant platform that handles inbound and outbound calls autonomously, allowing businesses to automate customer interactions. Sold as a white-label solution, it enables resellers to operate under their own brand. Joined as the founding engineer; built the entire AI layer from scratch and grew into project lead.

• Built the complete real-time voice AI pipeline from the ground up, integrating VoIP providers (Twilio, Telnyx) with AI services (OpenAI Realtime API, Gemini Live, ElevenLabs), and engineered latency optimization, silence detection (VAD tuning), and streaming audio handling for production-grade call quality.
• Designed a function-calling tool system using the registry pattern, enabling LLM agents to perform real-world actions during calls: forwarding calls to external numbers, transferring to live operators via WebRTC, creating reservations synced with Google Calendar, sending confirmation SMS, and placing orders from uploaded menus and catalogues.
• Built a RAG pipeline using Weaviate vector database, allowing businesses to upload documents and knowledge bases so AI agents could answer domain-specific questions during live calls.
• Designed and built a full outbound call center module with campaign management, enabling businesses to call thousands of contacts automatically — expanding the platform from inbound-only to a complete call center solution.
• Implemented Temporal for orchestrating outbound call workflows and automations, managing call scheduling, per-step timing of workflow actions, and configurable retry strategies based on user responses — ensuring reliable execution of long-running campaign processes at scale.
• Developed the multi-tenant white-label architecture with isolated databases per tenant and request-level domain resolution to dynamically route each incoming request to the correct database, allowing reseller partners to onboard and manage their own customers under custom domains and branding.
• Built backend infrastructure using Python (FastAPI), including RESTful APIs, PostgreSQL schema design with async SQLAlchemy, and Celery task queues for background job processing.
• Implemented CI/CD staging infrastructure using Docker Swarm and Traefik, with custom GitHub Actions workflows allowing developers to type /deploy on a PR to spin up isolated staging environments for real-time branch testing.
• Set up pgBackRest for reliable PostgreSQL backup and point-in-time recovery in Docker, along with Flower for Celery task monitoring.
• Served as project owner: planned timelines, assigned tasks, and led the development team while reporting directly to the CEO and management in planning meetings.

TAPSI is a ride-hailing company with more than 500,000 daily rides and 20 million users.

• Built and maintained scalable data pipelines using Python, SQL, and PySpark, enabling real-time data processing and performance optimization.
• Conducted exploratory data analysis and statistical evaluations to generate actionable insights.
• Developed a geographic search engine converting text addresses into coordinates, reducing third-party dependencies and achieving a 10% reduction in search time.
• Collaborated with cross-functional teams in an Agile environment, documenting methodologies and presenting clear technical reports to stakeholders.

• Designed and implemented a Retrieval-Augmented Generation (RAG) system that integrated vector databases to improve Large Language Model (LLM) context using LangChain, Weaviate, and AWS Bedrock.
• Developed end-to-end ML solutions using Docker and AWS, and adhered to CI/CD best practices.
• Authored technical documentation detailing model design, deployment processes, and performance evaluation.