// projects
Work
Systems I've designed and built — from distributed API gateways to multi-agent AI pipelines.
AI-Driven Podcast Generation
Event-driven pipeline processing long-form audio into synthesized podcasts. Fault-tolerant architecture handling high-concurrency workloads with async task orchestration, vector context engine, and multi-agent script generation.
FastAPIQdrantRedisAWSNext.jsPostgreSQLHuggingFace
500+ concurrent workflows
// deep dives
Architecture Breakdowns
How these systems work under the hood
NervaAI
AI-Driven Podcast GenerationSystem DesignEvent-DrivenAI/MLScalability
Async Task Orchestration
- ›Redis queues
- ›Job ID return pattern
- ›Celery workers
- ›FastAPI async
Vector Context Engine
- ›RAG pipeline
- ›HuggingFace embeddings
- ›Qdrant vector DB
- ›LangChain integration
Multi-Agent Script Generation
- ›Researcher + Writer agents
- ›GPT-4 backbone
- ›Custom agent orchestration
Optimized Asset Delivery
- ›S3 signed URLs
- ›CloudFront CDN
- ›Streaming responses
Challenges solved:
Long-running TTS (5-10 min)HTTP timeout managementMulti-agent state coordinationVector search for context retrieval
OpenSkill
AI-Powered Claude Skill OrchestratorCLILLMsGolangDeveloper Tools
Modular LLM Provider System
- ›Go interfaces
- ›Groq, OpenAI, Anthropic, Ollama support
Local State & Versioning
- ›.claude/skills/ directory
- ›SKILL.md rollback history
Skill Composition Engine
- ›Extends/includes support
- ›Template system
Challenges solved:
Skill format standardization across providersAsync AI generation in CLI contextCross-platform compatibility
VentaEdge
Distributed API GatewayDistributed SystemsGolangNetworkingSecurity
Intelligent Load Balancing
- ›Round-robin
- ›Least-connections
- ›Health checks
Distributed Rate Limiting
- ›Token bucket algorithm
- ›Redis + Lua atomics
Circuit Breaker
- ›Fail-fast on downstream errors
- ›Automatic recovery
Challenges solved:
Cascading failure preventionLow-latency overheadDistributed rate limit synchronization
SyncLayer
Real-Time Collaboration EngineReal-TimeWebSocketsDistributed SystemsGo
WebSocket Connection Layer
- ›Go Fiber stateful connections
- ›Connection pooling
Pub/Sub Message Bus
- ›Redis Pub/Sub across instances
- ›Fan-out pattern
Optimistic UI & Write Batching
- ›Batched PostgreSQL writes
- ›Client-side optimistic updates
Challenges solved:
Scaling WebSocket connectionsConcurrent edit conflictsDelivery guarantees
SentralQ
Multi-Agent AI DebuggingAI AgentsObservabilityLLMsAutomation
Log Ingestion & Normalization
- ›FastAPI intake
- ›Pandas processing
- ›Standard JSON schema
Agent Swarm Orchestration
- ›Supervisor + specialist agents
- ›LangChain backbone
Synthesized Solution Delivery
- ›Aggregated findings
- ›Reviewer LLM (GPT-4)
- ›Streaming output
Challenges solved:
Multi-agent coordinationHallucination controlVaried log format parsing