Building the internal AI stack for the marketing team: MCP servers, RAG chatbots, and a few multi-agent pipelines. Work that used to take an analyst two days now happens inside a prompt.
- Shipped a 0-tool MCP server wired to Google Search Console, GA4, Google Ads, HubSpot CRM, and Bing, so the team can pull marketing data from Claude Desktop without jumping between five dashboards.
- Designed the public support chatbot using hybrid BM25 plus dense retrieval on Pinecone with a Redis cache, embeddable as a one-line script. Answers land in under 0ms on the hot path.
- Built an ICP classifier that enriches cold company lists with Apollo and LinkedIn data, scores each prospect against Zenskar's $40K ACV B2B XaaS profile using a 120B Groq model, and pings sales on Slack when a high-fit lead lands.
- Prototyped a multi-agent competitor events scraper (Playwright, Apify, Groq, Gemini) that monitors 8 competitor sites and surfaces industry events every week. Cut about 4 hours of manual research from the team's schedule.
- Shipped the outreach tracker, a Streamlit-over-Supabase dashboard the SDR team now runs their entire weekly pipeline on.
PythonMCPLangChainPineconeRAGMulti-agent LLMGroqGemini