Suleman Manji

AI systems engineering, platform automation, agent runtime infrastructure, and public-safe operational proof by Suleman Manji.

Operator / Engineering Portfolio

MSP command planes that survive contact with real work.

I build the glue between service tickets, platform APIs, billing systems, agent runtimes, stakeholder artifacts, and proof gates. The work is less about demos and more about making messy operational state safe enough to act on.

View Work Map Featured Systems Contact
Eight-platform MSP tooling CWM execution control planes Microsoft Graph diagnostics AlgaPSA / vArida billing Qdrant / mem0 recall Stakeholder artifact pipelines
Current Operating Surface

The active work is a command center, not a portfolio theme.

Platform depthConnectWise Manage, CW Automate, ScreenConnect, Microsoft Graph, ITGlue, Sophos, Cavelo, vCloud, Meraki, Portainer, GitHub, and AlgaPSA/vArida all show up in the current local work surface.
Execution disciplineTickets become task structures, notes, status gates, schedule entries, time entries, and repo-side registers. Work is verified after writes instead of assumed complete.
Agent infrastructureSession logs, Project 4 items, Qdrant-backed memory, generated CLI contracts, and reusable skills are treated as operational infrastructure.
Public boundaryThe site abstracts client-sensitive details while keeping the engineering pattern visible: real platforms, real constraints, real proof.
Featured Systems

Evidence-led systems lanes.

MSP Operations Active

MSP Command Plane

Eight-platform MSP operations tooling across PSA, RMM, Microsoft 365, documentation, security, remote access, vCloud, and billing workflows.

Problem
Service work spans tickets, time entries, RMM state, Microsoft Graph evidence, project records, stakeholder artifacts, and approval gates.
System
A CLI and skill-based command plane that turns live platform state into reviewed actions, CWM task structures, GitHub Project 4 governance, and reusable handoff artifacts.
Proof
CWM ticket-control-plane registers, time-entry reconciliation artifacts, control-validate gates, live runtime checks, and public-safe session-harvest maps.
Evidence Signals
  • Eight TypeScript CLI packages and 250+ orchestration skills are represented in the local local-agent-tooling repo.
  • CWM tickets are treated as execution surfaces with tasks, notes, status, schedules, documents, and time entries reconciled after writes.
  • Internal development effort was reconciled against member 346 time entries, GitHub Project 4, git activity, browser history, and Codex session traces.
TypeScript Python GitHub Projects ConnectWise Microsoft Graph
Financial Agent Systems Research / Paper Trading

Financial-Agent Paper Runtime

Research infrastructure for paper-trading decisions, risk gates, reconciliation, forecasting, and runtime evidence.

Problem
Financial-agent systems need explicit safety boundaries, reproducible runs, and evidence before any operator trusts the output.
System
Paper-runtime components for decision policy, ledger checks, broker status reads, forecast scoring, and delivery gates.
Proof
Test suite results, paper-run logs, reconciliation checks, and configuration gates.
Evidence Signals
  • Research framing stays paper/runtime only, with safety language separated from operational MSP claims.
  • Runtime outputs are treated as evidence bundles rather than opaque recommendations.
  • Reconciliation and broker-state checks are part of the operating model before any downstream decision.
Python TypeScript Alpaca Paper Forecasting Risk Gates
TradingAgents-FinBots
Agent Runtime Active

Agent Memory and Runtime Infrastructure

Local agent runtime, memory, recall, and skill infrastructure that preserves operator context across interrupted sessions.

Problem
Agent work loses value when decisions, proof locations, tool state, and blockers disappear between sessions.
System
Qdrant-backed mem0 recall, checkpoint/replay helpers, session-harvest maps, repo-native skills, and explicit handoff prompts.
Proof
Hermes memory search results, Qdrant checkpoint guidance, session-log harvest artifacts, and validation notes tied back to GitHub control-plane items.
Evidence Signals
  • Local mem0 recovered Project 4 and local-agent-tooling effort history during this portfolio pass.
  • Session harvest maps convert raw Codex logs into skills, docs, CLI gaps, and Project 4 issue graphs.
  • Checkpoint wrappers are used when direct Qdrant writes are unreliable from sandboxed agents.
Python Qdrant mem0 Docker MCP
hermes-mem0
Active Lanes

Problem surface -> operating system -> proof path -> leverage.

MSP Operations MSP Command Plane CWM ticket-control-plane registers, time-entry reconciliation artifacts, control-validate gates, live runtime checks, and public-safe session-harvest maps. Financial Agent Systems Financial-Agent Paper Runtime Test suite results, paper-run logs, reconciliation checks, and configuration gates. Agent Runtime Agent Memory and Runtime Infrastructure Hermes memory search results, Qdrant checkpoint guidance, session-log harvest artifacts, and validation notes tied back to GitHub control-plane items. Operator Interfaces AlgaPSA and Operator Interfaces UIUX gap analysis, stakeholder YAML-to-HTML/XLSX and YAML-to-PPTX pipelines, runtime proof captures, and GitHub Project 4 issue gates. MCP and API Tooling MCP and Platform API Toolchain CLI contract generation, help-json surfaces, smoke tests, package builds, read/write gates, and public-safe investigation summaries.
Proof Stream

What counts as proof here.

CWM evidenceTicket state, tasks, notes, time entries, and post-write verification.
Runtime proofDeployed behavior, route checks, stack state, and smoke tests.
Control-plane gatesGitHub Project 4 fields, issue hierarchy, validation commands, and proof comments.
Memory continuityQdrant/mem0 recall, checkpoint wrappers, and session-harvest artifacts.
Audience Routes

Different readers need different proof.

MSP and service leadersOperational leverage, safer automation, and reviewable proof.
Engineering peersArchitecture, runtime boundaries, and tooling decisions.
Hiring managersScope, stack, ownership, and delivery evidence.
Agent/tooling collaboratorsMCP, memory, runtime, and workflow patterns.
Operating Principles

How I work.

  • Use the live system as the source of truth.
  • Preview risky writes and verify the result after execution.
  • Turn one-off troubleshooting into reusable skills, scripts, and runbooks.
  • Keep client-sensitive details out of public artifacts.
  • Make agent work durable enough for another operator to resume.