← Back home
Selected Work
A collection of projects spanning AI systems, realtime interfaces, and infrastructure-heavy deployments.
01
Private investigation operations platform
A private AI-assisted case workflow with local model serving, human review gates, and role-based approvals for sensitive legal operations.
“Reduced manual handoff time while keeping data inside controlled infrastructure.”
- Deployed local LLM inference with quantized models for confidential document processing
- Built multi-stage approval workflows with role-based access control
- Implemented audit logging and immutable operation trails
- Integrated vector search for case document retrieval
02
Live conversational avatar stack
A low-latency speech pipeline for a live avatar demo with memory-aware GPU scheduling and real-time voice interaction.
“Kept voice timing and avatar behavior stable during sustained event traffic.”
- Orchestrated STT → LLM → TTS pipeline with sub-2s end-to-end latency
- Built WebSocket-based streaming with backpressure handling
- Implemented session memory management for long conversations
- Optimized GPU scheduling for variable traffic patterns
03
Industrial edge vision network
Computer-vision workloads near industrial sites with lightweight MQTT telemetry back to operator dashboards.
“Enabled realtime alerts without depending on cloud video streaming.”
- Containerized CV services with ONNX runtime for edge deployment
- Built MQTT broker infrastructure for reliable telemetry ingestion
- Implemented anomaly detection with threshold alerting
- Designed dashboard for multi-site operator visibility