AI ENGINEER
2025
← HOME

Working Philosophy

The principles that guide my development process

01

AI-First Thinking

Start with the intelligence layer. Route tasks to the right model based on complexity, latency, and cost. Use local inference for sensitive data, cloud for scale.

02

Iterate in Production

Ship fast, measure everything. Real user feedback beats perfect planning. Deploy daily with automated rollback. Monitor latency, error rates, and user satisfaction in real-time.

03

Build to Scale

Start with serverless, add infrastructure when needed. Design for horizontal scaling from day one. Use managed services until you need control, then self-host strategically.

04

Automate Everything

If I do it twice, I automate it. Workflows for deployment, testing, monitoring, and data processing. Let AI handle the repetitive, keep humans in the loop for judgment.

Reasoning Engines

The LLMs I route to based on task requirements

COMPLEX REASONING
CLOUD

GPT-4o

OpenAI · 128K context · $5/1M tokens

My go-to for multi-step workflows, code generation, and strategic planning. Strong tool-use capabilities and reliable structured output.

When: Complex logic, code generation, multi-step tasks
DOCUMENT ANALYSIS
CLOUD

Claude 3.5 Sonnet

Anthropic · 200K context · $3/1M tokens

Best for long-context analysis, detailed reasoning, and instruction following. Superior safety and nuanced understanding of complex documents.

When: Large documents, careful analysis, compliance requirements
FAST TASKS
CLOUD

GPT-4o-mini

OpenAI · 128K context · $0.15/1M tokens

Cost-optimized for classification, summarization, and routine automation. 15x cheaper than GPT-4o with ~300ms latency.

When: High-volume tasks, simple classification, rapid prototyping
PRIVATE DATA & FINE-TUNING
SELF-HOSTED

Llama 3.1 70B + QLoRA

Ollama Runtime · Unsloth Fine-Tuning

Local inference for sensitive data. Custom fine-tuned models utilizing QLoRA adapters for domain-specific security intelligence and OSINT data processing. Zero external API calls, complete privacy control.

When: QLoRA model tuning, enterprise compliance, custom offline workflows
VOICE AI
HYBRID

ElevenLabs + Whisper

Conversational AI · <800ms latency

Real-time voice interactions with Whisper STT, LLM reasoning, and ElevenLabs TTS. WebRTC streaming for sub-second responses.

When: Customer service, voice assistants, accessibility features

Development Stack

Daily tools and frameworks I work with

TERMINAL
LOCAL

Warp Terminal

Rust-based · AI-native

GPU-accelerated terminal with AI command generation, workflow blocks, and collaborative sessions. My command center for everything.

When: Every day, all day — command execution, debugging, deployment
FRONTEND
FRAMEWORK

React + Vite

React 18 · Tailwind CSS

Modern React with Vite for instant HMR. Tailwind for rapid UI development. Component-driven architecture with TypeScript for type safety.

When: Dashboards, admin panels, customer-facing interfaces
BACKEND
SERVERLESS

Vercel Edge Functions

Node.js · Global CDN

Serverless functions at the edge. <50ms cold starts, automatic scaling, deployed across 30+ regions. Perfect for API endpoints.

When: API routes, webhooks, serverless backends
DATABASE
CLOUD

Supabase

PostgreSQL 15 · Real-time subscriptions

Managed Postgres with automatic API generation, row-level security, and real-time subscriptions. Edge functions for serverless compute.

When: Transactional data, user management, real-time features
VECTOR SEARCH & RAG
SELF-HOSTED

Weaviate + Advanced RAG

Hybrid Search · Cohere Rerank · Semantic Cache

Vector database and retrieval orchestration. Implementing lexical-vector hybrid search, cross-encoder rerankers, and custom chunking strategies to minimize LLM hallucinations.

When: High-scale semantic search, context retrieval, document Q&A
CONTAINERS
ORCHESTRATION

Docker Swarm

v24.x · 30+ services

Container orchestration for microservices. Service discovery, load balancing, rolling updates, and automatic failover across 6 nodes.

When: Microservices architecture, self-hosted infrastructure
CLOUD INFRASTRUCTURE
HYBRID

Google Cloud Platform

Vertex AI · Cloud Run · BigQuery

Deploying scalable AI applications and microservices. Utilizing Vertex AI for enterprise model deployment, testing, and managed dataset orchestration.

When: Enterprise model pipelines, containerized scaling, GKE deployments
EVALUATION & HARNESS
LOCAL

Agent Test Harnesses

LLM-as-a-Judge · Benchmark suites

Custom evaluation harnesses for testing agent behavior, reliability, and security vulnerability discovery. Automating prompt injection tests and tracing latency profiles.

When: Validating agent safety, grading model output consistency, regression testing

Automation & Integration

How I connect systems and automate workflows

DURABLE EXECUTION ENGINE
SELF-HOSTED / GO & PYTHON

Temporal.io

50+ Go/Python workflows · 380K+ activities/day

Resilient, code-defined orchestration with event-driven state tracking. Implements auto-retry mechanisms, complex sagas, and asynchronous callbacks with strong consistency guarantees.

When: High-throughput ingestion queues, agent coordination DAGs, billing & lifecycle logic
SaaS INTEGRATION
CLOUD

Zapier Premium

80+ Zaps · 50K monthly tasks

Enterprise SaaS integration for CRM, email, calendar, and business tools. Multi-step automation with conditional logic.

When: Connecting SaaS tools, marketing automation, business workflows
PROTOCOL BRIDGE & MIDDLEWARE
SELF-HOSTED

MCP Server & Middleware

Model Context Protocol · Dynamic Tool Bindings

Standardized middleware for LLM-to-system communication. Custom JSON-RPC tool endpoints, prompt registries, and context-injection hooks for autonomous agents.

When: Connecting LLMs to local databases, filesystem access, agent tool calling
AGENTIC ORCHESTRATION
FRAMEWORK

LangGraph & CrewAI

Stateful cyclic graphs · Multi-agent systems

Orchestrating autonomous agents with state preservation, custom tool bindings, conditional routing, and human-in-the-loop checkpoints.

When: Multi-agent automation, self-correcting reasoning loops

Quality Standards

The benchmarks I target for production systems

<200ms
API Latency
99.7%
Uptime Target
Daily
Deployments
100%
Test Coverage
<5min
Rollback Time
Auto
Scaling