Selected Work
By Paul Tomanpos, Jr.
AI-Powered Work Log
Intelligent statuses and next Steps for Concurrent projects
Transforming Fragmented Voice-Notes into Actionable Next Steps, Mitigating Context Switching with Intelligent Automation
Responsibilities
✓
System Architect
✓
UX Designer
✓
Technical Integrator
✓
Prompt Engineer
Overview
- Designed and built a functioning prototype to streamline work logging across multiple concurrent projects.
- Eliminates repetitive manual context switches by leveraging AI-driven summaries and structured task workflows.
- Created a system architecture integrating iOS Shortcuts, Zapier, Google Sheets, and OpenAI API.
- Personally deployed and validated in daily workflows, demonstrating smoother task management and reduced overhead.
Problem Framing
The Challenge of Context Switching
The Critical Pain Point
The core problem was the severe cognitive friction and productivity loss caused by managing numerous diverse projects simultaneously, including fabrication, website development, and operational tasks. Restarting work required significant mental effort to remember precisely: “Where did I leave things?” and “What is the next 5-minute, 30-minute, or 2-hour task I must execute to make progress?”.
Why Traditional Logging Failed
Manual logging methods often introduced unnecessary friction and distraction:
- Context Switching Overhead: Logging notes on a computer meant opening multiple browser tabs, leading to distractions and sidetracking.
- Fragmentation: Notes were often scattered across various apps, making consolidation and synthesis of project status nearly impossible.
The need was for a dedicated, voice-first “assistant of sorts” that could quickly capture status updates and instantly provide intelligent, actionable summaries to maintain project continuity.
| Before: Manual Logging | After: AI Work-Logging |
| Step 1: Open Note-taking App Step 2: Open/Search for File Step 3: Hand-Type Entry Step 4: Add Tags/Categories Step 5: Save / File It | Step 1: Tap Project Icon Step 2: Speak (Auto-Saves) |
Manual logging required multiple steps per work log entry. With AI summarization, the same log is captured in essentially two steps: Tap then Speak.
Solution
The AI-Powered Voice Log System
I designed and built an end-to-end automated workflow that transforms simple voice commands into structured project data and intelligence.
Core Technology Stack
The solution showcases systems design for multi-tool integration, leveraging several commercial and proprietary tools:
| Layer | Component | Function |
| Input/UX Layer | iOS Shortcuts (Voice) | Captures voice input, triggers automation, and delivers audible summaries. |
| Orchestration Layer | Zapier (Webhooks) | Serves as the central automation engine, routing transcribed data between iOS and the database. |
| Database Layer | Google Sheets | Stores the centralized, chronological, time-stamped work log for all active projects. |
| Intelligence Layer | Openrouter API | Leverages advanced Generative AI models (e.g., Anthropic, GPT-4, Gemini) for processing, interpretation, and structured output generation. |
Optimized UX Flow 1: Instant Voice Logging
The logging process was engineered for speed, prioritizing minimal friction and single-tap input.
- Project Initiation: The user taps a dedicated project icon (e.g., “Faux Tree”) located on the iOS home screen.
- Voice Input: The user speaks their progress update (e.g., “Painted base layers period”).
- Automatic Categorization: The Shortcut automatically captures the voice transcription, generates timestamps, and embeds the correct project identifier (a three-letter abbreviation) associated with the specific icon used.
- Data Ingestion: The data is sent via a webhook to Zapier, which instantly appends the entry to the designated Google Sheets database.
This flow ensures accurate project categorization by design, eliminating the need for manual tagging or sorting.
Optimized UX Flow 2: AI-Powered Summary Retrieval
The retrieval flow transforms raw notes into actionable project intelligence.
Summary Request: The user taps a single shortcut icon labeled “Project Summary,” then chooses the project from a list.
AI Synthesis: Openrouter queries the Google Sheet for all entries for that project. Utilizing advanced AI prompt engineering, the system interprets the chronological notes to generate a structured output.
Audible Delivery and Archiving: The response is delivered audibly (essential for hands-free workflow). Once the summary finishes speaking (or is stopped by the user), the text response is “loaded” as a text message to the user, allowing for optional archiving and later reference.
The structured AI output always includes three critical components:
- Summary of current status,
- To-Do List (in order of urgency), and
- Next Steps
Challenges & Strategic Solutions
Overcoming technical and financial constraints was crucial for validating this complex prototype.
| Challenge | Mitigation / Solution | Key Skill Demonstrated |
| AI Token Consumption Costs | The potential cost risk of using high-end commercial models (like GPT-4) was mitigated by leveraging Openrouter. I strategically switched to more token-efficient LLMs (e.g., Anthropic) that could reliably perform the “lower-weight processing” task of summarizing data, ensuring the prototype remained fiscally sustainable. | Strategic Resource Optimization |
| Latency and Data Overload | To preemptively ensure the system felt “fast” and reliable, I intentionally contained the scope, relying only on 1–2 sentence work logs. This controlled the input size and retrieval payload, stabilizing the database queries and AI processing time. | Architectural Constraint Avoidance |
| iOS Development Friction | While the system required extensive setup in the cumbersome iOS Shortcuts mobile interface, this development friction was accepted as a necessary trade-off for quickly validating the optimal user experience (single-tap entry). | Prioritization of UX over Dev Convenience |
Results & Impact
Though this was a personal project without formal user testing, I actively used the system in my own daily multi-project workflows. This provided practical validation of the design’s effectiveness and ensured the architecture functioned as intended.
Minimizing Context Switching: The computer-free, voice-first input mechanism completely avoids the distraction caused by navigating desktop tabs and apps, enabling the user to maintain focus on the task at hand and “make some progress on it”.
Guaranteed Project Continuity: The system replaces the time-consuming effort of manually searching fragmented notes by delivering an instant, actionable summary structured with the “Next Steps” required to transition back to any of the 6–8 active projects.
Hands-Free Decision Making: The inclusion of audible output allows the user to consume complex project summaries while physically working (e.g., fabrication tasks), supporting a seamless, hands-free, multitasking workflow.
Validation of Novel Architecture: The successful integration of disparate platforms (iOS, Sheets, Zapier, Openrouter) validates the underlying architecture for building reliable, multi-tool AI agents.
Next Steps & Key Learnings
This prototype served as a critical validated learning experience for designing human-centered AI systems.
Key Learning
The project demonstrated that a robust and efficient AI workflow can be created by stitching together existing, reliable tools (iOS, Zapier, Google Sheets) and leveraging AI orchestrators (Openrouter) to perform intelligent data processing. The result is akin to having a “conversation with an assistant who is keeping track of the progress of all the things I’m working on”.
Future Vision: Automated Semantic Task Reconciliation
The immediate priority for the next iteration is to evolve the system from a summarizer to an intelligent agent capable of automated task reconciliation.
This would involve engineering the AI layer to:
- Semantically Track Completions: Analyze subsequent log entries (e.g., “I did it a week later”) to link back to previously identified To-Do list items.
- Manage Project State: Automatically mark tasks as completed within the database, transitioning the system to actively manage project state, further reducing user effort and maintaining up-to-date summaries.
This refinement would fully realize the potential of workflow automation by applying GenAI capabilities to project management, a key area of innovation in the technology sector.
Reflections & Lessons Learned
- Balancing automation with user control is crucial — AI speeds up routine tasks, but trust requires transparency and fallback plans (Future Improvements).
- While no formal user research was conducted, personal validation demonstrated the core mechanics were sound. In future iterations, involving teams would stress-test the system in collaborative settings.
- Designing for AI requires both system-level architecture and human-centered interaction — each reinforces the other.