MOHAMMED
SALIH_
Lead GenAI Engineer / Agentic AI Architect / MCP Specialist
I build governed agent systems, production MCP servers, RAG pipelines, and secure tool-use workflows.
Lead engineer for governed agentic AI systems
|
I specialize in LLM systems that retrieve enterprise context, call tools safely, and expose decisions clearly enough for operators to trust. My recent work covers production MCP servers, hybrid RAG pipelines, LangChain and AutoGen orchestration, evaluation harnesses, guardrails, RBAC allowlists, and Docker-first delivery.
Capabilities
A senior AI portfolio should show what kind of problems can be trusted to you. These are the capability areas I repeatedly deliver across agentic AI, retrieval, reliability, product, and backend systems.
Agent Architecture
Designs plan-execute workflows, ReAct loops, tool registries, memory patterns, and human checkpoints for autonomous systems.
MCP & Secure Tool-Use
Builds governed Model Context Protocol servers that expose enterprise tools through typed contracts, RBAC, allowlists, and sandboxed execution.
RAG & Knowledge Systems
Creates retrieval pipelines over structured and unstructured data with query routing, metadata enrichment, deduplication, reranking, and source grounding.
LLM Reliability
Implements evaluation harnesses, regression checks, guardrails, PII filtering, permission boundaries, and traceable execution paths.
AI Product Engineering
Turns AI capability into product features with clear workflows, operator visibility, mobile/web interfaces, documentation, and stakeholder alignment.
Backend Delivery
Ships production APIs and AI microservices with Python, FastAPI, Docker, SQL databases, MongoDB, Node.js, and deployment documentation.
“A model becomes useful when it can touch the world safely,
remember context, inspect evidence, and ask another system to act.”
Selected Work
A portfolio should make the work scannable. Each case study here shows the problem, my role, the engineering approach, and the business or reliability outcome.
Enterprise MCP Tool Layer
Zealogics / Governed AgentsEnterprise AI services needed controlled access to SharePoint, SQL databases, and workflow triggers without giving LLMs unrestricted system access.
Architected and deployed four production MCP servers with typed JSON Schema contracts, RBAC allowlists, and sandboxed tool execution.
Enabled secure governed tool access across organization-wide AI services.
Autonomous Agent Orchestration
Plan-Execute SystemsLong-running agent workflows needed planning, recovery, and operator checkpoints instead of brittle one-shot LLM calls.
Engineered LangChain plan-execute DAGs, ReAct reasoning, Human-in-the-Loop checkpoints, and self-correction loops.
Reduced agent failure recovery time by 50%.
Enterprise RAG Pipelines
Retrieval & ReliabilityFragmented enterprise documents needed citeable, reliable retrieval for production AI features.
Built hybrid retrieval with pgvector, BM25, Pinecone, FAISS, metadata enrichment, query routing, Cohere reranking, and deduplication.
Reduced hallucination rates to below 2% with grounded context management.
Aurora Hypercore / plugLLM
Open Research & SDKsAgent experiments and provider integrations needed reusable abstractions for memory, tool registries, provider switching, and fallback orchestration.
Created experimental repositories for autonomous agent workflows and unified LLM provider abstraction.
Produced reusable patterns for hierarchical execution loops, hot-swapping providers, and load-balanced fallback flows.
Evidence
Senior portfolio content needs proof. This section collects CV-backed outcomes, role evidence, published thinking, certifications, and awards.
Built production MCP servers exposing enterprise tools through typed contracts, allowlists, and controlled LLM tool access.
Developed AutoGen and LangChain agent workflows for business process automation, structured outputs, persistent memory, and multi-step document processing.
Led the bridge from traditional Python and React Native product lines into GenAI features, Dockerized AI microservices, and stakeholder-aligned AI roadmaps.
Let's Build
Bring a complex workflow, a stalled AI prototype, or an enterprise system that needs agentic intelligence. I can help architecture, prototype, harden, and ship it with clear controls.