Kalder Knowledge Assistant

A RAG-powered Q&A assistant grounded in the Kalder Personalization Hub corpus — a synthetic nine-document framework describing a B2B buying group personalization system for a fictional enterprise platform. Created for portfolio purposes: all company names, team members, and business data are fictional.

What it does

Ask a question and get an answer grounded exclusively in the Kalder corpus, with a source citation for every claim. If the corpus doesn't cover something, the assistant says so explicitly rather than guessing.

Tech stack

• Streamlit for the interface
• Pinecone for vector storage
• OpenAI text-embedding-3-small for embeddings
• Claude claude-sonnet-4-6 for answer generation

Setup

1. Install dependencies:

   pip install -r requirements.txt
   

2. Add your API keys to .streamlit/secrets.toml:

   ANTHROPIC_API_KEY = "..."
   OPENAI_API_KEY = "..."
   PINECONE_API_KEY = "..."
   

3. Run the app:

   streamlit run knowledge_assistant.py
   

4. In the sidebar, click Rebuild Index to ingest the ten Kalder corpus files (nine markdown documents plus the data model) into the pm-knowledge Pinecone index. This flushes any existing vectors first.

Using the app

• Ask — ask a question grounded in the Kalder corpus and get a cited answer with retrieved sources, token usage, and estimated cost.
• Ingest Log — see what's currently indexed: chunk count and last-ingested timestamp per document.
• Eval — run an 8-question golden-set smoke test across all nine corpus documents to catch retrieval or index regressions.

About the knowledge base

The Kalder Personalization Hub corpus is a synthetic nine-document framework describing a B2B buying group personalization system for a fictional enterprise platform. It covers role architecture, signal definition, audience segmentation, content modeling, decisioning rules, journey convergence, measurement, operations, and privacy. It was built for portfolio purposes — there is no real company called Kalder.