Founder & CEO at Magnence — an AI & Software Development Company building intelligent products from architecture through production. I work across the full stack of modern AI systems: generative AI and LLM applications, RAG pipelines, agentic workflows, and the cloud/DevOps/LLMOps layer that keeps them reliable once they ship.

I care less about knowing every framework and more about shipping systems that hold up in production — observable, secure, and cost-conscious, not just demoable.

Generative AI apps · LLM systems · RAG pipelines · AI agents & agentic workflows · multi-model orchestration · embeddings & vector search · AI-powered SaaS

Backend systems & REST APIs · full-stack apps · microservices · database architecture · auth & authorization · API integrations · developer tooling

Containerized apps · cloud-native architecture · CI/CD pipelines · production deployment · infra automation · monitoring & observability

Model integration & prompt pipelines · RAG evaluation · guardrails & confidence thresholds · observability · multi-provider LLM architecture

AI-powered developer tooling that analyzes GitHub repositories to understand architecture, project structure, data flow, workflows, documentation, dependencies, and development patterns.

AI Code Intelligence
Repository Analysis Architecture

Privacy-first, self-hosted job search orchestrator — multi-platform discovery and tracking without handing data to a third party.

AI Agents Automation
Self-Hosted FastAPI PostgreSQL

Edge-AI adaptive traffic signal system using YOLOv8 and Webster's Algorithm for real-time optimization.

Computer Vision YOLOv8
Edge AI Real-Time Systems

• vLLM — Accept Anthropic tool addition/removal content blocks
• FastAPI — Support Annotated in stream return annotations for JSONL and SSE
• LlamaIndex — Fix Ollama streamed tool call deduplication
• Haystack — Type-check agent hooks tool result offloading (Merged)
• RAGFlow — Support recent Gemini models integration (Merged)
• BentoML — Working directory path sanitization in service imports
• Docling & Unstructured — Document markdown emphasis & OCR loader error handling

• Qdrant — Fix DBSF reversing ranking order for Euclidean distance
• Chroma — Fail fast on empty embeddings validation
• Ragas — Preserve caller kwargs on metric model creation
• DeepEval — Validate HumanEval evaluation params
• Guardrails AI — Fix on-fail handler naming mismatches & schema detection
• Promptfoo — Image-to-Video AI custom provider example

• OpenTelemetry — Fix flaky BatchProcessor shutdown test (Merged)
• Langfuse — Type metadata, config, and modelParameters in OpenAPI schemas
• Google ADK — Pre-populated session events in SessionInput for eval cases
• Google ADK Recipes — Cloud deployment instructions & scaffold path
• Gemini CLI — Terminal ghost text wrapping & A2A server routes
• n8n — AI Agent tool call empty content string normalization

• Ruff — Clarify isort settings in RUF022 documentation
• Pytest — Document set_log_path() usage and hook examples
• Metaflow — Decompress list IndexError guard on empty strings
• OpenVINO — PyTorch frontend aten::ravel operator support
• PEcAn — Package builds warning handlers (Merged) & sensitivity analysis

DISCOVER

Define the problem, users, constraints, requirements, and success criteria.

ARCHITECT

Design the system architecture, services, APIs, data flows, and infrastructure.

DATA

Build reliable data pipelines, knowledge sources, schemas, embeddings, and storage.

INTELLIGENCE

Integrate models, prompts, agents, RAG, retrieval, tools, and orchestration.

VALIDATE

Apply guardrails, evaluation, confidence thresholds, testing, and human oversight.

SHIP

Deploy, observe, optimize, secure, and continuously improve the production system.