Venkata Sai Kumar Vendoti

I build and maintain the AI systems behind Fn7’s go-to-market platform, including agents, integrations, retrieval pipelines, and LLM orchestration.

• Brain agent: Built a 25-tool agent that creates a knowledge graph of a company’s products, target audience, competitors, and brand assets. Content, campaign, and SEO features use this information to produce more consistent, brand-aligned outputs. Rebuilt the agent using Google ADK; the updated version runs in production.
• Agent Builder: Designed and built the frontend and backend of a no-code platform for creating agents through pathways and stages. It supports server-side callbacks, automatic prompt generation, and autonomous execution. I also implemented a meta-agent that generates agent definitions and prompts.
• MCP integrations: Developed a custom MCP server that exposes 12 platform APIs as tools, including LinkedIn, X, Reddit, Instagram, Apollo, and an internal knowledge graph. Agents use these tools to interact with external platforms and retrieve relevant brand context on behalf of users.
• LLM orchestration: Built an LLM router supporting six providers, including OpenAI, Google Gemini, Anthropic, and AWS Bedrock. It handles provider fallbacks, tracks token usage and cost per call, and uses an LLM-as-judge step to evaluate outputs.
• Hybrid RAG: Designed a retrieval pipeline for SEO keyword-coverage analysis using BGE embeddings and BM25. Combined the results using reciprocal-rank fusion, then reranked them with a cross-encoder. A calibration step converts model scores into covered or not-covered decisions.
• AI audit optimization: Reduced the cost of an AI audit by approximately 40×, cutting LLM calls from around 120 to 4 while maintaining per-question citation accuracy. Also improved latency and memory usage through embedding-model singletons and content-hash caching.
• Video generation: Built the Video Studio generation pipeline using Google Veo 3.1 and ElevenLabs voice cloning. Each new clip continues from the previous clip’s last frame to help maintain character consistency across multiple clips.
• Full-stack and infrastructure: Led the frontend migration from Angular to Next.js, work on CI/CD and Kubernetes infrastructure, and managed a team of three interns on client projects.

Worked on a veterinary AI application, document and voice processing, and frontend form infrastructure.

• Built a veterinary AI platform independently. It extracts pet medical history from documents using Google Cloud Vision and converts voice-based treatment notes into visit summaries and discharge reports using Speech-to-Text.
• Developed custom JSON Forms renderers for grid, flex, and array layouts, and fixed a synchronization issue so form data updates across pages.
• Fixed a cross-platform audio issue affecting Chrome on Windows and macOS.