Engineering leader building reliable agentic systems, AI guardrails, and practical developer tools.
Director at Orion Photo Group. Building Zohal Systems.
I lead engineering delivery across enterprise platforms, web applications, integrations, and AI-assisted workflows.
My current focus is reliable AI delivery: getting useful speed from agents without giving up verification, permission boundaries, observability, or human accountability.
I build and document the operating layer around agentic systems:
- context and specifications that keep agents aligned
- verification checkpoints that catch confident failures
- guardrails that constrain production blast radius
- delivery systems that connect AI output to real business workflows
- planning methods that account for review, testing, rework, flow, and risk
I am also building Zohal Systems, an AI-augmented engineering studio focused on practical delivery systems for web applications, CRM, integrations, and AI-assisted workflows.
Production-focused safety hooks for Claude Code that intercept destructive commands before execution.
- PreToolUse guardians for AWS, Salesforce, and databases
- escalating controls based on the cost of being wrong
- passkey-gated confirmation for production writes
- installer, configuration examples, demo, and fail-closed tests
- explicit limitations and compatibility with least-privilege access
This project explores a simple principle: agent permissions should reflect the blast radius of an incorrect action.
A focused Claude Code hook that limits runaway sub-agent spawning before it drains a session quota.
- cumulative per-session spawn limits
- optional estimated token budget
- concurrency-safe session counters
- Python standard library only
- explicit fail-open behavior for malformed hook payloads
The project treats uncontrolled agent fan-out as an operational boundary, not only a model-quality problem.
An open framework for estimating, grooming, and calibrating AI-assisted engineering work.
AIDEM separates the work AI may accelerate from the work teams still own:
- context and requirements
- human review
- testing and validation
- rework
- delivery flow
- production and customer risk
It includes transition, AI-native, and hybrid planning modes, plus templates, risk scorecards, pilot guidance, and remediation playbooks.
A Python desktop toolkit for merging, splitting, rotating, and converting PDF documents.
The project demonstrates conventional software-delivery discipline alongside AI-focused work:
- modular application structure
- automated tests
- build and release workflows
- packaged Windows releases
- checksum verification
- security, support, roadmap, and contributor documentation
A Windows-first voice workflow for AI prompts using local Whisper or Groq, with dedicated Urdu and English input paths.
PromptPilot Voice is a public-source developer preview. Its public license remains pending upstream clarification, so the repository is available for learning, review, and transparency rather than reuse or redistribution.
- reliable agent orchestration and bounded autonomy
- context engineering and durable agent memory
- verification systems for AI-assisted delivery
- AI governance implemented in the execution path
- Salesforce and enterprise-system integration
- engineering leadership in AI-augmented teams
- multilingual and local-first developer workflows
Artifacts over claims. A repository, test, demo, or measured result is more useful than a broad prediction.
Guardrails before autonomy. More agent freedom should follow stronger context, verification, permission boundaries, and recovery paths.
Verification is part of delivery. Review and testing are not cleanup after AI generates an answer. They are part of the system that makes the answer usable.
Humans retain accountability. Agents can compress execution, but people still own judgment, exceptions, and production outcomes.
Design for failure and recovery. Before asking whether an agent can complete a task, ask how it can fail, who will know, and whether the action can be stopped or reversed.
I write about reliable agentic systems, AI-assisted engineering delivery, guardrails, verification, and the operating changes engineering leaders face as AI enters real workflows.