Founder of TrueLogics · 9+ years shipping for teams in the US and Europe
Your AI demo works. Getting it onto real data, permissions, workflows, and costs is the hard part. That's the part I build.
I build AI agents, permission-aware RAG, MCP servers, and the SaaS and mobile products around them, from architecture through production. You work directly with me: the person designing the system and writing the code.
- Production AI, not demos. Permissions, retrieval, cost controls, tracing, and deployment are part of the work.
- The AI and the product. Agents and RAG plus APIs, databases, web, mobile, and infrastructure. One engineer owns the stack.
- AI-assisted, human-reviewed. I use Claude Code and Cursor to move faster, and I review every change that ships.
| StudioX AI | Enterprise AI that grounds assistants in a company's own documents and runs multi-step operational workflows. Multi-tenant SaaS and on-prem. |
| Vectra24 | Coordinated Marketing, Sales, and Support agent teams that small businesses direct from the chat apps they already use. |
| Pivot | Collaboration platform with docs, messaging, and audio/video on web, iOS, Android, and desktop. About 20 Go/TypeScript services, plus MCP and AI features. |
| ClosePinger | Client portal for bookkeeping firms: requests files, chases them automatically, and tracks month-end close. |
| ProntoMigo | Delivery marketplace for food, grocery, retail, van transport, and transfers, with live tracking and AI ops for vendors. |
| EveryCare | Senior-care coordination for individuals, families, and professional caregivers. Offline-first Expo app. |
| RidePlus | Rides, parcels, groceries, cargo, and freight on one dispatch and one fleet. |
| OptimalLab | Interactive explainers for tokens, attention, RAG, evals, agents, system design, and DSA, with practice problems. |
| Luman | macOS storage app that shows what's using disk and explains why each cleanup is safe. |
➡️ Case studies at muhammadsarmad.com/work
🧠 Engineering OS — @truelogics
A platform that gives AI coding agents an organization's actual engineering decisions — its ADRs, rules and architecture — instead of general best practice. Built in Go, designed as a kernel with independent consumers.
| engineering-kernel | The kernel and the eng CLI — collect, parse, index, retrieve, classify. SQLite + FTS5, no embeddings, no vector DB, by design. |
| engineering-mcp | An MCP server that exposes that knowledge to Claude Code, so a review can cite the decision it contradicts. |
| engineering-review | The first consumer: pull-request review grounded in retrieved context rather than a diff in isolation. |
| engineering | The rulebook the platform reads — standards, ADRs, RFCs, and the policy governing what gets built. |
Developed through documented RFCs and controlled experiments — including one that measured why an AI client silently ignored a registered MCP server, and found the cause was deferred tool schemas rather than the tool descriptions everyone assumed.
Go · SQLite/FTS5 · Model Context Protocol · JSON-RPC · 9 RFCs
Agents & tool calling · Permission-aware RAG · Hybrid search + reranking · Evals & tracing · Agentic workflows
Hourly · Full-time · Part-time · Project-based (milestones or modules)
We pick the mode, scope, and terms on a short discovery call. No surprise scope. You own all the code and IP.
Available for new projects. Send what you're building: an agent, RAG on your docs, MCP into your app, or the product around it.