I help teams take AI prototypes and data-heavy platforms to production-grade Go and cloud architectures.

I’ve shipped high-impact systems at Microsoft, AWS, Property Finder, and DataGardens – from Editor AI in Windows Photos (+20% MAU) and RDS SQL Server features, to subscription/credit platforms that cut support tickets by 50%+ and DR/HA platforms that made cutovers predictable instead of stressful.

Today I run AandZ.tech, an independent AI & cloud engineering practice based in the UAE, working with global clients on AI integration, distributed systems, and event-driven architectures.

Production AI & Platforms

• Turn Python/PoC AI pipelines into reliable Go/ cloud services (GenAI, RAG, agents, vector search).
• Design and harden event-driven systems (CQRS, outbox, idempotent consumers, replayable projections).
• Make AI features fast, observable, and affordable (local-first where possible, cloud where it counts).

Consulting & Architecture

• Fractional Principal Engineer / Architect for teams scaling from MVP to “we need this to not fall over.”
• Architecture reviews for fintech, marketplaces, and SaaS: consistency, invariants, and scaling plans.
• Mentoring engineers on microservices, observability, and AI integration.

• Fractional Principal Engineer (Go + Cloud)

  • Design / review event-driven and microservices architectures.
  • Focus: scalability, fault tolerance, and clear service contracts.

• RAG & Vector Search Optimization

  • Fix latency / quality issues in Qdrant / pgvector pipelines.
  • Hybrid search (keyword + vector), re-ranking, and cost/latency tuning.

• Python → Go Migration

  • Rewrite or peel off hot paths from Python/Node backends into Go services.
  • Goal: higher concurrency, lower cloud spend, cleaner deployment story.

• Agentic Systems & Orchestration

  • Design deterministic agent workflows (plans, tools, state, observability).
  • Genkit / LangChainGo patterns, structured outputs, and guardrails.

If you’re a CTO, founder, or lead engineer wrestling with any of these, my repos below are essentially reference architectures for the conversations we’ll have.

Repo: github.com/amirhf/creditLedger • Story: From Spreadsheet Chaos to Reliable Credits

A production-ready reference architecture for handling high-volume financial credits and balances. It replaces fragile "credits column" designs with an immutable, double-entry ledger that guarantees auditability.

• Event-Driven Integrity: Implements the Transactional Outbox pattern and idempotent consumers to ensure consistency across microservices without dual-write hazards.
• CQRS & Scalability: Decouples write logic from read patterns using CQRS projections, allowing for high-throughput ingestion and replayable read models.
• Full Observability: Pre-wired with Jaeger and Grafana to trace "what happened to my credit?" journeys and monitor consumer lag/latency.

Use it as a blueprint for: SaaS credits, marketplace wallets, usage-based billing, and loyalty point systems.

Repo: github.com/amirhf/imageSearch

A high-concurrency Retail Discovery Engine combining Go for search orchestration and Python for ML inference.

• Polyglot Architecture: Uses Go on the hot path (search/fusion) for sub-100ms latency and Python for the rich ML ecosystem (CLIP/BLIP).
• Hybrid Search: Merges Dense Vectors (Qdrant) and Sparse Keywords via Reciprocal Rank Fusion (RRF) to solve the "zero results" problem in e-commerce.
• AI Feature Router: A "circuit breaker for intelligence" that cuts cloud costs by ~90% by dynamically routing requests to local quantized models before falling back to cloud APIs.
• SaaS Scalability: Features built-in multi-tenancy, Shadow Mode for safe migration validation, and full OpenTelemetry instrumentation.

Use it as a reference for: e-commerce discovery, hybrid search implementation, and migrating Python prototypes to Go systems.

Repo: github.com/amirhf/learningPathDesigner

A production-ready reference architecture for building deterministic AI agents in a distributed system. It replaces fragile "chatbots" with a structured planner-executor workflow grounded in domain data.

• Planner–Executor Pattern: Implements a multi-stage agentic workflow where a Planner proposes the learning path and an Executor decomposes it into concrete steps with verified resources.
• Structured Outputs: Solves non-deterministic LLM behavior by enforcing strict JSON schemas for all plans and quizzes, ensuring reliability for the UI.
• Advanced RAG & Re-ranking: Features a production-grade retrieval pipeline using Qdrant (hybrid search) and Postgres to ground AI responses in real documentation.
• Polyglot Microservices: Demonstrates a scalable layout: Go Gateway (orchestration & auth) managing Python AI services, ensuring distinct separation of concerns.

Use it as a blueprint for: Building robust internal training platforms, onboarding copilots, and domain-specific certification agents.

Go · Python · TypeScript / React / Next.js
FastAPI · Postgres · MySQL · SQL Server · MongoDB
Qdrant · pgvector · Kafka / Redpanda · RabbitMQ
AWS (Kinesis, Glue, Athena, Step Functions, RDS, S3) · Azure
Docker · Kubernetes · Grafana · Prometheus · Jaeger · App Insights

• 🌍 Based in Dubai (UTC+4) — working remote / global.
• 🧩 Ideal fit: teams that already have users and data, and now need their AI / platform to be reliable, observable, and scalable.
• ✉️ Best way to reach me: check my email / links in the sidebar or visit AandZ.tech.

If one of my reference architectures looks close to what you’re building, that’s usually the fastest way for us to start a concrete conversation.