Daily recap for June 21, 2026

Today we shipped 7 new step-by-step tutorials that small businesses can download, run locally, and adapt to their own operations. Each solution pairs production-grade AI with a specific business problem—from cutting Databricks LLM costs to automating chargeback responses for Stripe merchants.

New tutorials

Databricks AI Spend Control for Budget-Conscious SMBs

LLM costs can spiral unpredictably when agents handle varied traffic. This solution wraps every Databricks LLM call with real-time token-level spend tracking and automatically downgrades to cheaper models or halts requests when an agent hits its daily budget. The result: small businesses running Databricks AI agents can enforce hard limits without manual oversight.

Read the tutorial → · Download the code (zip)

Built with @reaatech/agent-budget-engine, @reaatech/agent-budget-middleware, @reaatech/llm-cost-telemetry-calculator, @reaatech/llm-router-engine, @reaatech/agent-budget-spend-tracker on Databricks · 103 tests / 95.15% coverage.

Databricks Lead Intake Agent for SMB Financial Advisors

Independent financial advisors lose too many web leads to slow manual sorting. This agent classifies incoming messages, extracts contact details even from unstructured text, enforces per-lead budget caps, and hands off qualified prospects to HubSpot for scheduling—all through a single API endpoint backed by Databricks LLMs.

Read the tutorial → · Download the code (zip)

Built with @reaatech/confidence-router, @reaatech/session-continuity, @reaatech/structured-repair-core, @reaatech/agent-handoff, @reaatech/agent-budget-engine on Databricks · 107 tests / 98.16% coverage.

Google Gemini Medical Claim Extraction for SMB Practices

Small medical practices spend too many hours rekeying data from faxed or scanned claim forms. This pipeline uses Gemini to extract patient demographics, diagnosis codes, and billing items from PDFs, repairs the JSON against a strict schema, redacts PII, and caps per-claim processing costs. It's built as a Next.js API with async job processing and Supabase storage.

Read the tutorial → · Download the code (zip)

Built with @reaatech/structured-repair-core, @reaatech/guardrail-chain, @reaatech/llm-cost-telemetry, @reaatech/agent-budget-engine, @reaatech/session-continuity on Google Gemini · 96 tests / 98.62% coverage.

Mistral AI Code Sandbox with Approval Gates for SMB Data Teams

Business analysts want to run analytical Python scripts from natural-language prompts, but executing untrusted code is risky. This solution forces every generated script through a human approval gate before it executes in an isolated E2B sandbox, with Mistral generating the code and a Next.js dashboard showing requests pending review. Budget caps and full audit trails come standard.

Read the tutorial → · Download the code (zip)

Built with @reaatech/tool-use-firewall-core, @reaatech/agent-handoff, @reaatech/agent-budget-engine, @reaatech/llm-cost-telemetry, @reaatech/session-continuity on Mistral · 165 tests / 96.20% coverage.

Vertex AI Voice Agent for Buildium Maintenance Requests

Property managers miss after-hours calls from tenants with emergencies. This 24/7 voice agent answers Twilio calls, authenticates tenants against Buildium, and uses Vertex AI Gemini to create or update work orders—transcribing speech with Deepgram and responding with Cartesia TTS. It replaces manual after-hours logging and keeps tenants informed.

Read the tutorial → · Download the code (zip)

Built with @reaatech/voice-agent-core, @reaatech/voice-agent-telephony, @reaatech/session-continuity, @reaatech/voice-agent-stt, @reaatech/voice-agent-tts, @reaatech/llm-cost-telemetry on Vertex AI · 162 tests / 96.84% coverage.

Perplexity Security Guardrails for QuickBooks Online Invoice Processing

SMBs want to automate invoice classification but fear exposing vendor or customer PII to AI models. This guardrail layer intercepts every Perplexity call, masks personal data, blocks prompt injection, enforces financial policy rules, and logs all decisions to Langfuse. It lets accounting teams safely use AI for invoice categorization and anomaly detection.

Read the tutorial → · Download the code (zip)

Built with @reaatech/guardrail-chain, @reaatech/guardrail-chain-guardrails, @reaatech/guardrail-chain-config, @reaatech/llm-cost-telemetry, @reaatech/structured-repair-core, @reaatech/guardrail-chain-observability on Perplexity · 113 tests / 96.69% coverage.

Databricks Agent Mesh for Stripe Dispute Response Automation

E-commerce merchants lose chargeback disputes because they can't gather evidence and rebut in time. This agent mesh coordinates specialist agents: one pulls transaction data from Stripe, another drafts the rebuttal using a Databricks-hosted LLM, and a confidence router files automatically or escalates to a human reviewer. It preserves multi-step state in Upstash Redis and ships with a dashboard for manual overrides.

Read the tutorial → · Download the code (zip)

Built with @reaatech/agent-mesh, @reaatech/agent-mesh-router, @reaatech/agent-mesh-session, @reaatech/agent-mesh-observability, @reaatech/confidence-router, @reaatech/agent-handoff on Databricks · 90 tests / 100.00% coverage.

Browse all solutions →

Browse the full catalog at reaatech.com/products.