Muhammad Ali

• Architected Mera Brain, an autonomous ReAct-pattern AI reasoning loop with dynamic tool-calling (synapse_lookup,
knowledge_search, place_order) and a 5-turn execution cap, eliminating infinite LLM loops.
• Enforced zero-hallucination pricing by limiting the AI agent to SKU/quantity extraction, resolving all prices from
a PostgreSQL catalog through Gos service layer (shopspring/decimal) for financial accuracy.
• Designed a 4-service gRPC/Protobuf modular monolith — ERP Core, Synapse (Identity Resolution), Knowl
edge (pgvector search), and Mera Brain — as the platforms core backend architecture.
• Enforced multi-tenant isolation at the database layer using PostgreSQL Row-Level Security and transaction
scoped context (SET LOCAL app.current_business).
• Built full AI observability— tracing prompts, reasoning steps, tool calls, and retrieved chunks to Grafana/Prometheus
— to monitor LLM token usage and RAG latency.

• Automated AWS packaging/licensing pipelines in Node.js — spinning up Windows EC2 instances on demand,
running build scripts, uploading installers to S3, and self-terminating — cutting infrastructure idle costs by
70%.
• Built a cross-platform PDF and annotation manager in Electron/TypeScript with a scale-aware coordinate
system that kept annotations aligned across 0.25x-3.0x zoom levels.
• Engineered a progressive Browser Cache API caching layer, cutting load times for 50-100MB PDF files by
80%.
• Integrated FFmpeg to automate local audio compression of book media, reducing desktop bundle sizes.

• Built a full-stack ERP/CRM system from scratch (MERN stack) to run inventory, sales, and customer lifecycle
management as the companys only engineer.
• Built an autonomous Gemini-powered customer support chatbot, increasing qualified sales leads by 25%.
• Cut LLM API costs 30% by building a pgvector semantic caching layer into the RAG pipeline.
• Enforced multi-tenant isolation via PostgreSQL Row-Level Security, keeping customer data cleanly separated.