A public-safe portfolio repository for a multi-agent voice + messaging system focused on practical AI application engineering.
This showcase highlights how I approach:
- multi-agent orchestration
- voice pipeline integration (STT → LLM → TTS)
- model routing and fallback design
- task lifecycle tracking
- delivery reliability and cost awareness
It is intentionally sanitized for public release: no production credentials, no private user data, no raw internal prompts, and no proprietary workflow logic.
This is not a chatbot toy repo.
It is a showcase of the system layer around AI applications: how requests are routed, how agents coordinate, how long-running work is tracked, how voice fits into the flow, and how reliability / cost tradeoffs are handled in practice.
- explicit role boundaries
- structured handoff patterns
- final-delivery ownership
- internal coordination without duplicate user-facing output
- inbound voice handling
- speech-to-text normalization
- text generation
- text-to-speech output
- channel-aware delivery with text fallback
- task-based model selection
- premium / default / utility / fallback tiers
- balancing quality, latency, availability, and cost
- degraded paths under provider failure or rate-limit pressure
- append-only task status events
- stale-task awareness
- delivery dedupe and cooldown handling
- long-running workflow visibility
User Message / Voice
→ Ingress Layer
→ Speech-to-Text (if voice)
→ Context Assembly
→ Router
→ Agent Workflow / Handoff
→ Model Invocation
→ Response Assembly
→ Text-to-Speech (optional)
→ Delivery Layer
→ Status Ledger / Monitoring
This repository focuses on the middle engineering layers: routing, orchestration, reliability, and public-safe packaging.
See also:
openclaw-agent-showcase/
├── README.md
├── LICENSE
├── .gitignore
├── .env.example
├── docs/
├── examples/
├── scripts/
├── config/
└── diagrams/
docs/— architecture and engineering notesexamples/— redacted interaction and task-flow examplesscripts/— small public-safe demo utilitiesconfig/— example-only configuration filesdiagrams/— source diagrams for repo visuals
This repository may include:
- redacted agent handoff examples
- redacted voice-session examples
- example task lifecycle events
- demo routing and cost-estimation utilities
- sanitized config templates
This repository does not include:
- production API keys
- real user data
- raw private session history
- private prompt bodies
- proprietary strategy logic
- live deployment configs
See docs/sanitization-boundary.md for the publication boundary.
- Python
- LLM APIs / model routers
- speech-to-text
- text-to-speech
- messaging platform integration
- SQLite task / status tracking
- workflow scripting and automation
- docs-first architecture communication
The hard part of AI products is usually not the first model call.
It is the system around that call:
- making multiple components cooperate
- deciding which model should handle what
- preserving context without wasting cost
- managing retries and degraded paths
- keeping delivery reliable
- making the system observable and maintainable
That system layer is what this showcase is meant to highlight.
This is a sanitized showcase repository, not a full production dump.
If you are reviewing this repository from a hiring perspective, the main signals are:
- system design thinking
- workflow reliability awareness
- AI integration pragmatism
- architecture communication quality
- engineering judgment around safe public sharing