
Throughline
Builds a knowledge brain over your real engineering work, then drafts posts in your voice paired with on-brand generated images. Multi-model routing across Vertex AI, Gemini, Grok and Midjourney. Nothing auto-posts — you stay the editor.
I build production AI systems — RAG pipelines, agentic workflows, and MCP tooling for Claude Code and Cursor — with Next.js, Python and Go on AWS.
Senior AI Engineer and full-stack lead with 5+ years shipping AI-powered web products to production — RAG pipelines, agentic workflows, LLM integrations (Claude, OpenAI, Groq) and MCP tooling, on AWS with Next.js, Node.js, Python and Go. I build my own tools in the open: ContextZero, a VS Code extension for managing what every AI coding agent knows about your projects, and PresenceZero, a self-hosted lead finder — both MIT-licensed. At Nextpak Agile Solutions I grew from developer to senior engineer and team lead, delivering end to end for 20+ international clients and mentoring the engineers around me. I'm Top Rated on Upwork with a 100% job success score.
Nextpak Agile Solutions
Freelance / Independent

Builds a knowledge brain over your real engineering work, then drafts posts in your voice paired with on-brand generated images. Multi-model routing across Vertex AI, Gemini, Grok and Midjourney. Nothing auto-posts — you stay the editor.

A free site of 39 tools and about 450 pages for developers building with LLMs: real-tokenizer token counts, daily-refreshed API prices, GPU/VRAM checks, Claude Code config builders and verified error fixes. Runs in the browser for $0 a month.

Self-hosted lead finder for freelance devs and agencies. Scan a location and category; it flags businesses with no site, a dead site or a parked domain, checks matching domain availability via RDAP, and scores each lead 0–100.
One VS Code sidebar for the rules, memory and skills every AI coding agent knows about each project — Claude Code, Cursor, Kiro, Antigravity and the AGENTS.md standard, with monorepos nesting properly.

A social operating system for agencies: deploy and monitor AI personas across client accounts on Twitter, LinkedIn and Reddit. Routes tasks across Claude, GPT and Gemini, with human approval gates before any action fires.

AI hiring and career platform spanning web, mobile and browser. Runs live AI video interviews, places automated phone screens, and ships a Chrome extension that auto-applies on LinkedIn, Indeed and Glassdoor.

No-code builder for AI chat and voice agents. Train on your PDFs, docs and site, then deploy to a web widget, WhatsApp, Messenger and Instagram — or to a phone line that books consultations into Google and Microsoft calendars.
Nextpak Agile Solutions · 2024
Recognized for multi-stack expertise, strong problem-solving, and handling complex technical challenges.
Nextpak Agile Solutions · 2025
Awarded for technical excellence, reliability, and consistent delivery of scalable solutions.
Virtual University of Pakistan
2020 – 2024 · Punjab, Pakistan
Tested fixes and deep dives from real projects — Claude Code, MCP servers, RAG pipelines and Next.js.
Codex compacts a long conversation automatically at 90% of the model's context window: 244,800 tokens on the 272K models in Codex 0.166.0. Where that number comes from, what model_auto_compact_token_limit can and can't change, why there's no off switch, how to compact between turns instead of mid-task, and what to do when compaction loops or loses your task.
Codex has no setting called auto mode. The closest thing is Approve for me (Auto-review): the same sandbox as the default, with a reviewer agent approving what you used to approve. How to turn it on in the CLI, config.toml and the app, what it reviews, why it still asks or denies, what replaced --full-auto, and how it compares with Claude Code's auto mode.
Claude Code says 'Not logged in · Please run /login' when the session has no credential it can use, and 'Login expired · Please run /login' when renewing your claude.ai login failed and the login was cleared. Since 2.1.318 there is a third: 'Login reached its time limit'. What each one means, how to check when your login ends, why Claude Code keeps logging you out, and how to stay signed in for scripts with claude setup-token.