Prototyping & Building AI Tools to Leverage Knowledge Work

Developing agentic systems that augment human capabilities through human-in-the-loop AI.

AI Leverage Diagram

Current Prototypes

Real Estate Investment Analysis

partnering

Agent evaluates investment opportunities using financial analysis and market data synthesis.

financial analysismarket researchmcp

Stack:

  • • Framework: Google ADK
  • • Model: Gemini 3 Pro
  • • Tooling: MCP, Web Search, Custom Tool (Python)
  • • Backend: GCP Cloud Run
  • • Inference: Vertex AI
  • • Frontend: Web App

Personal Tax Return Preparation

development

Agent assists in preparing tax returns by guiding collection and validating information.

tax analysisrag / vectorization

Stack:

  • • Framework: LangChain
  • • Model: Kimi-K2-Thinking
  • • Tooling: RAG, Custom Tool (Python)
  • • Backend: GCP Cloud Run
  • • Inference: Fireworks
  • • Frontend: Web App

Household Groceries Management

Ideation

Voice-first agent for household grocery management and inventory tracking.

agentic commercevoice

Stack:

  • • Framework: Google ADK
  • • Model: Gemini Live
  • • Tooling: MCP, Web Search
  • • Backend: GCP Cloud Run
  • • Inference: Vertex AI
  • • Frontend: Android App

Prototyping Process

Ideation

1

Brainstorming and validating AI agent concepts that solve real operational pain points through domain research and expert interviews.

  • • Identify high-friction manual processes
  • • Map stakeholder workflows
  • • Define success metrics
  • • Assess technical feasibility

POC Development

2

Building functional proof-of-concepts to validate technical approach and demonstrate core capabilities with human-in-the-loop validation.

  • • Select technology stack
  • • Build minimal viable agent
  • • Integrate domain-specific tools

Partnering

3

Collaborating with domain experts and early adopters to refine the solution through real-world testing and co-development.

  • • Identify design partners
  • • Onboard & train users
  • • Collect usage analytics
  • • Iterate based on feedback

MVP Development

4

Building a production-ready MVP with enhanced features, robust error handling, and scalable deployment architecture.

  • • Implement production features
  • • Add monitoring & logging
  • • Scale infrastructure
  • • Prepare for wider audience

Tooling

Agent Frameworks

Custom agent orchestration systems for complex workflow automation with human oversight controls.

  • • Structured agent deployment
  • • Human-in-the-loop checkpoints
  • • Multi-agent coordination

Knowledge Integration

Systems that synthesize domain-specific knowledge with real-time data and expert validation.

  • • Domain model construction
  • • Multi-source synthesis
  • • Expert validation workflows

Augmentation Patterns

Interface patterns that enhance rather than replace human expertise through intelligent assistance.

  • • Co-pilot interfaces
  • • Suggestion & refinement
  • • Transparency & explainability

Building Custom Agentic Systems for Maximum Impact

Agentic systems transform minimal human effort into massive impact. This is the power of AI in knowledge work when precisely integrated with human expertise.

Systematic analysis of existing workflows identifies areas for AI leverage, and process redesign defines where to implement agentic workflows for reliable, high-volume tasks. Human expertise is focused mostly on complex, high-value decisions and final quality checks.

An intelligent mixture of augmentation and control creates a superior, more productive, and reliable process than either humans or agents could achieve alone.

The Agentic Edge

Manual Process8 hours
Human + Agent30 minutes
+90% time savings+15x leverage

Creators

Diego Navarro

20-year career focused on high-impact roles in Fintech, E-commerce, Strategy Consulting, and Corporate Development. Leveraging a robust academic foundation in AI/ML, Finance, and Engineering.

AgenticSandbox Founder