Projects / Applied AI engineering
← All projectsVoice is the interface.
The system is everything behind it.
I built five connected voice-agent engineering projects covering API orchestration, typed tools, identity and authorization, evaluation, event delivery, observability, and mobile integration. A conversation becomes an engineering problem as soon as it can mutate a customer record.
Engineering focus
Conversational interfaces
API orchestration
Security & reliability
iOS / Android integration
ML inference boundaries
Five connected projects
One system, five hard problems.
Conversation → verified tool call
Voice Agent Workflow Orchestration
I built a voice-agent workflow gateway that turns appointment requests into authenticated API operations, with tenant-scoped access, idempotent updates, and recovery when a response disappears after a commit.
Conversation state · JSON tool contracts · Authorization · Idempotency
Architecture and implementationRegression tests · Bad calls on purpose
Voice Agent Evaluation & Red-Team Harness
I built a repeatable agent-evaluation harness covering caller verification, injected instructions, tool permissions, missing inputs, interruptions and human escalation. Good manners are not a security policy.
Deterministic eval fixtures · Guardrail checks · Regression evidence
Architecture and implementationLatency · Dependencies · Error budgets
Voice Agent Reliability & Operations
I implemented an agent operations reference with request counters, P50/P95 latency distributions, modeled dependency outages, bounded retry budgets and circuit-breaker behavior. A convincing conversation is only half the system.
Node.js · Request counters · Seeded P50/P95 simulation · Timeouts · Circuit breakers
Architecture and implementationWebhooks · Idempotent writes · Reconciliation
Event-Driven CRM Integration
I implemented a voice-to-CRM event service with HMAC-verified webhooks, replay-window checks, tenant-scoped idempotency, duplicate detection and recovery after simulated downstream outages.
REST/JSON · Webhooks · Tenant isolation · Outbox patterns
Architecture and implementationiOS · Android · Secure callback requests
Voice Agent Mobile Integration
I developed Swift and Kotlin callback clients, a secured first-party API contract, and Core ML / TensorFlow Lite inference adapters. The backend owns consent checks, case authorization, and provider-facing operations.
Swift · Kotlin · URLSession · Coroutines · Core ML/TFLite adapters · Backend APIs
Architecture and implementationWhere trust changes
Speech-to-text and LLM outputs are proposals. A typed tool gateway, authenticated principal and backend transaction decide whether anything actually changes. The model proposes; the authenticated backend decides.
Built across the stack
I built the Node.js API integration, signed CRM event receiver, evaluation fixtures, failure models, and native Swift/Kotlin callback clients. The implementations have executable HTTP contracts and regression tests. The public case studies omit private provider configuration, customer data and unpublished model artifacts.