GenAI · RAG01
Enterprise RAG Reference
Most RAG demos answer anything, cite nothing and ignore who is asking.
1.00hit@4 retrieval
0RBAC leaks · 17-case eval
1.00correct refusals
PythonHybrid BM25 + vectorRRFFastAPIChromaRBAC
View on GitHub →Open source
Built in
the open.
Five reference projects for GenAI and agentic AI, built the way an enterprise has to run them: grounded, governed, measured and cheap to operate. Every number below comes from a real run in the repo, and each one runs free on a laptop with no API key.
01
Projects
Five repos, real results.
Hover or tap a demo to play it. Recorded from each project's own CLI output.
Agentic AI · LangGraph02
Presales Copilot Agents
Responding to an RFP takes days of reading, mapping, sizing and risk analysis.
12/12requirements traced
296.2person-days estimated
< 1 sper RFP
LangGraphMulti-agentPythonOllama / OpenAI / Anthropic
View on GitHub →Agentic AI · Data03
Lakehouse Text-to-SQL Agent
Letting an LLM write SQL against a warehouse risks wrong numbers and unsafe statements.
0.79execution accuracy
8/8attacks blocked
~9 msper question
DuckDBsqlglotSemantic layerSnowflakeDatabricks
View on GitHub →MCP · Governance04
Enterprise MCP Toolkit
MCP makes tools trivial; making them safe, scoped and auditable is the hard part.
3governed MCP servers
15security tests passing
$0run cost
Model Context ProtocolDuckDBPydanticAudit logClaude / Cursor
View on GitHub →LLMOps · Evaluation05
Agent Eval Harness
Teams ship prompt and model changes to agents based on vibes.
4regressions caught
29%cheaper, still blocked
exit 1CI gate
YAML suitesLLM-as-judgeSQLiteStreamlitCI
View on GitHub →Want one of these in your stack?