Motivational Communication assistant

Inspiration

Managers often know what outcome they need, but difficult feedback can easily become directive, accusatory, or demotivating. I want to build an AI-powered Ethical Management Communication Coach: a toolset that helps managers remain clear about expectations while protecting autonomy, competence, and trust.

This assistant is the first part of that broader toolset. I started it as a personal project four days ago, before I knew about OpenAI Build Week. I learned about the hackathon only a few hours before submission and used that time to extend the prototype with an English browser experience for judges.

What it does

The assistant transforms a harsh or overly directive managerial message into constructive communication without hiding deadlines, decisions, or accountability.

For each message, it provides:

• a short diagnosis of the communication problem;
• a direct but developmental alternative;
• a metaphor-based alternative;
• an explanation of the changes;
• a Self-Determination Theory check covering autonomy, competence, and relatedness.

It also contains an ethical refusal path for requests intended to humiliate, deceive, or manipulate someone. The goal is not generic tone polishing, but theory-grounded management communication.

How we built it

The application uses GPT-5.6 for structured message analysis and transformation. Codex helped me design and implement the Core API, prompt contracts, retrieval pipeline, feedback flow, deployment configuration, and browser interface.

The technical implementation includes:

• FastAPI Core API;
• GPT-5.6 structured output;
• multilingual embeddings and RAG;
• pgvector retrieval of approved communication examples;
• a versioned communication playbook;
• feedback and regeneration controls;
• audit and model-cost tracking;
• Telegram and browser adapters;
• Supabase PostgreSQL and Railway deployment.

The browser interface is intentionally simple: no signup, a sample message, one clear action, and a coherent before/after result.

Codex sessions used:

• Primary Build Week session: 6a5ffbde-55f4-83ed-b33e-6b0f1b0b7308
• Initial project session: 019f6fbc-d40b-73f2-a3b4-1499f4cfbbf0

Challenges we ran into

The main challenge was turning communication theory into enforceable product behavior rather than a vague “make this friendlier” prompt. Each generated alternative must first acknowledge something truthful about the person and only then propose a direction, while preserving real managerial constraints.

Another challenge was that I joined the hackathon only shortly before the deadline, I had to add a browser interface and isolated public deployment too quickly without proper testing, evals setting and UI/UX improvement.

Accomplishments that we're proud of

I am proud that I was able to produce such a complex logic and deployment with Codex with just several hours spent from my personal time. The project is already a working end-to-end system rather than only a prompt demonstration. It combines GPT-5.6, retrieval, structured output, feedback, ethical safeguards, rate limiting, and audit tracking.

The assistant makes a specific case for real impact: helping managers communicate difficult feedback without destroying employee motivation or trust.

What we learned

I learned that reliable AI communication support requires more than a strong model. It needs an explicit methodology, carefully selected examples, structured output validation, ethical boundaries, and a feedback loop.

I also learned how effectively Codex can accelerate a project across architecture, implementation, debugging, deployment, and UX—while leaving the important product and ethical decisions with the builder.

What's next for Motivational Communication Assistant

This is intended to become a broader Ethical Management Communication Coach. The next steps are:

• additional tools for feedback, delegation, conflict, and change communication;
• more languages and culturally appropriate examples;
• integrations with workplace communication tools;
• operational excellence improvement including evals, prod monitoring, etc.
• stronger privacy controls, retention policies, and human review workflows.

Built With

• api
• codex
• css3
• docker
• embeddings
• fastapi
• gpt-5.6
• html5
• javascript
• openai
• outputs
• pgvector
• postgresql
• pydantic
• python
• rag
• railway
• rest
• structured
• supabase