Cornelius Byas

# Portfolio Case Studies ## Case Study 1: Automated Lead Qualification with Zapier - **Client:** Early-stage fintech startup struggling to keep up with incoming leads from their Typeform signup. - **Challenge:** Sales reps were spending 2–3 hours daily manually reviewing, categorizing, and routing each new lead—leading to delayed follow-ups and lost opportunities. - **Solution Flow:** 1. **Trigger:** New Typeform submission 2. **Action:** “Webhooks by Zapier” sends the form data to OpenAI’s ChatGPT API with a prompt to score lead quality (e.g. “On a scale of 1–10, how qualified is this prospect based on their answers?”) 3. **Filter:** Zapier filters out any lead scored below 5 4. **Action:** High-scoring leads automatically get: - A new contact created in HubSpot CRM - A Slack notification posted to the sales channel - A row appended to a “Qualified Leads” Google Sheet - **Tools Used:** Zapier, Typeform, Webhooks by Zapier, OpenAI API, HubSpot, Slack, Google Sheets - **Outcome:** - **80% reduction** in manual lead triage time - **Immediate follow-up** on hot leads via Slack alerts - **20% boost** in MQL-to-SQL conversion rate within the first month ## Case Study 2: Context-Aware Knowledge Assistant with LangChain - **Client:** Mid-sized software company needing an internal “smart help desk” to surface documentation snippets and keep context across follow-up questions. - **Challenge:** Employees bounced between Confluence pages and chat threads, losing conversation context and frequently asking repeat questions. - **Solution Architecture:** 1. **Ingestion:** Imported ~5,000 Confluence pages into a FAISS vector store 2. **Agent Chain:** Built a LangChain `ConversationalRetrievalChain` that: - Retrieves the most relevant docs for the current query - Feeds them, plus the conversation history from a Redis-backed memory buffer, into OpenAI’s GPT-4o model - Returns a concise, context-aware response 3. **Interface:** Deployed as a lightweight Flask app with a React frontend and a Redis store for session memory - **Tools Used:** LangChain, FAISS, Redis (for memory), Python (Flask), React, OpenAI API - **Outcome:** - **60% faster** resolution of internal support queries - **0 repeat questions** within the same session thanks to memory persistence - **95% employee satisfaction** rating in a post-launch survey