zeeshan g.

I've architected and shipped production AI systems running across multiple countries, not demos. Among them: PropAI, a multi-tenant, Claude-powered WhatsApp AI agent platform on Meta's Cloud API serving real-estate agencies across Africa; CropSight, a GAN-based computer-vision system for precision agriculture; and an agentic financial-intelligence platform built on LangGraph + FastAPI. PhD in Computer Science (specialization: Machine Learning, AI and Computer Vision), 15 years in AI, former CTO. I help startups and enterprises turn LLM capabilities into reliable, deployable architecture (retrieval, memory, tool use, orchestration, voice, evaluation, and scalable cloud or on-premise infrastructure), not prompt-engineering experiments that fall over in production. SYSTEMS I'VE SHIPPED PropAI: multi-tenant WhatsApp AI agent platform for real-estate agencies and agents (Meta Cloud API, Embedded Signup, Coexistence). Claude-driven LLM orchestration via n8n, with RAG + agentic responses on agents' live numbers. Deployed across African markets on region-isolated, data-residency-compliant AWS; property-market feasibility studies delivered for UK clients. Enterprise code-intelligence platform (in delivery): architecting a fully on-premise, air-gapped AI system for a large enterprise-software firm. Self-hosted open-weight LLM (vLLM) with hybrid RAG and a deterministic code-dependency engine, serving a 22-developer team under strict data-residency constraints that rule out cloud APIs. Agentic financial-intelligence platform: LangGraph orchestration over structured financial data, generating personalized audio content end-to-end (LLM + TTS). CropSight: computer-vision platform for precision agriculture and corporate-farm compliance. Fuses satellite, drone, and mobile imagery with GAN-based super-resolution mapping low-resolution sources to high-resolution field detail. Deployed across African markets. Elena AILL (research collaboration): advising on architecture approach, technical direction, and MVP build for a from-scratch, non-transformer persistent-memory AI research project. Enterprise AI & telematics integration for PepsiCo, plus AI architecture and code audits for international clients. WHAT I BUILD End-to-end RAG pipelines (hybrid retrieval, reranking, vector DBs) Agentic & multi-agent systems with tool use, memory, and orchestration Voice AI: TTS, voice-to-voice / voice conversion, and production voice-agent pipelines Self-hosted and on-premise AI: vLLM, open-weight models, full local stacks with monitoring, for data-sovereignty and cost-sensitive environments Fine-tuning and LoRA/QLoRA pipelines Persistent-memory and continual-learning systems Workflow automation at production scale (n8n, self-hosted) Scalable inference infrastructure and cloud-native ML systems on AWS CI/CD and MLOps workflows for AI products LLM evaluation and performance-benchmarking frameworks STACK ML/DL: PyTorch, TensorFlow, Hugging Face Transformers, scikit-learn, OpenCV · LLMs: Claude, GPT, Llama, Mistral, Qwen, open-source models · Serving: vLLM, self-hosted / on-premise, quantization (AWQ, GPTQ) · Orchestration: LangGraph, LangChain, LlamaIndex, n8n (production, self-hosted), custom · Voice/Speech: TTS, voice-to-voice (RVC), Whisper STT · Vector/DB: Qdrant, pgvector/Supabase, Pinecone, FAISS, BGE reranker · Backend: FastAPI · Cloud: AWS (EC2, EKS, S3, SageMaker, IAM, monitoring) · Deployment: Docker, Kubernetes · WhatsApp Business / Meta Cloud API Most developers build AI demos. I architect systems with clear technical decisions, cost-awareness, evaluation built in, and production-readiness from day one, and I work at the research edge of memory and continual-learning architecture, not just integration. If you're building an AI-native product and need architectural clarity, scalability strategy, and senior technical leadership, I can structure and execute it. I also do fixed-scope AI architecture and code audits, a fast, low-risk way to start.