# Ashad Abdullah Product Engineer, AI, Founder · Islamabad ## For founders and hiring managers Ashad Abdullah is a product engineer and founder who ships AI products, agents and plugins end to end, in whatever stack the problem needs. Open to founding and product engineer roles, and taking freelance builds. - Owns work end to end: founded Fyltr and took it from idea to launch solo; shipped ImagineArt's plugins into five host apps from spec to release; led an agentic video editor from prototype to first release in about 8 weeks. - Any stack: picks the tools after the problem; projects range from assembly to Rust. - AI-native: runs multiple coding agents in parallel workspaces, at the speed of thought, and owns what ships. - Proof of work: every claim below links to a shipped product, case study or repository. - Taking freelance builds · open to full-time roles. Contact: ashad001sp@gmail.com I'm Ashad Abdullah, a product engineer who builds AI products and agents for founders and teams, from idea to launch. Shipped ImagineArt into Premiere Pro, After Effects, Photoshop, Figma, Framer and Chrome, built a 120+ tool agent for traders that answers in ~30s instead of 200s, and took Fyltr from idea to 2.3K signups on a zero-budget launch. - 5: Production plugin surfaces - 120+: Tools orchestrated in one agent - 2.3K: Signups on a zero-budget launch ## Services - AI inside the tools people already use: Plugins, browser extensions and MCP servers that put your product inside Premiere Pro, Figma, Chrome or an AI agent, built for each host's real runtime and distribution. ([ImagineArt plugins case study](https://ashad001-portfolio.vercel.app/work/imagineart), [Motif, an MCP server](https://ashad001-portfolio.vercel.app/work/motif)) - Agents and LLM systems: Multi-agent systems with tool routing, streaming and evaluation built in. TradersGPT routes across 120+ tools, streams its answers and is traced end to end. ([TradersGPT case study](https://ashad001-portfolio.vercel.app/work/tradersgpt), [Bolt-SQL, text-to-SQL in Rust](https://ashad001-portfolio.vercel.app/work/bolt-sql)) - Products from idea to launch: The whole product: the app, the AI underneath and the launch. Fyltr went from an idea to an AI form builder with document extraction and 40K+ visitors in launch week. ([Fyltr case study](https://ashad001-portfolio.vercel.app/work/fyltr), [The Fyltr launch post](https://www.reddit.com/r/ClaudeAI/comments/1rr47ya/i_delayed_my_product_launch_for_months_because_i/)) ## Availability Taking freelance builds · open to full-time roles. Based in Islamabad, Pakistan (UTC+5). ## Contact - [Send a brief](https://ashad001-portfolio.vercel.app/#brief): what you're building, who it's for, and when you need it - [Email](mailto:ashad001sp@gmail.com) ## Work - [ImagineArt](https://ashad001-portfolio.vercel.app/work/imagineart): Putting a generation engine everywhere the work already happens Associate PM → PM, surface strategy owner · ImagineArt (Vyro) · Jun 2026 — Present · Shipped, in production Shipped into 5 host apps and a Chrome extension, on a platform with millions of users. People don't want to visit your agent. They want their agent to use your product, without leaving the tool they already have open. - [TradersGPT](https://ashad001-portfolio.vercel.app/work/tradersgpt): 120+ tools, and a routing layer that decides between them Senior AI Engineer — architecture owner · TraderWare · Jul 2025 — Jun 2026 · In production 120+ tools behind one agent, with time to answer cut from 200s to ~30s. Past a certain number of tools the model stops choosing well and starts guessing. The routing layer matters more than the tools do. - [Fyltr](https://ashad001-portfolio.vercel.app/work/fyltr): Founded it, launched it, and failed to charge for it Founder — solo engineer, solo marketer · Independent · Jul 2025 — May 2026 · Paused 40K+ visitors in launch week and 2.3K signups, with no budget. I've repeatedly built things people wanted and failed to charge for them. Closing that gap is most of what I'm working on now. ## Lab - [Bolt-SQL](https://ashad001-portfolio.vercel.app/work/bolt-sql) (2025): Text-to-SQL that answers before you look away - [Motif](https://ashad001-portfolio.vercel.app/work/motif) (2026): An MCP server that watches your screen recording and finds the bug - [Devhire](https://ashad001-portfolio.vercel.app/work/devhire) (Mar 2023 — May 2023): A behavioral interview chatbot, built before that was an obvious product - [claude-arcade](https://github.com/Ashad001/claude-arcade) (2026): Rust + ratatui terminal Minesweeper wired into Claude Code lifecycle hooks, auto-launches in a tmux split - [Cue](https://cue.fyltr.co) (2026): Hyperlocal event and hackathon discovery agent, topic-lock primitive, two-pass scoring model - [WebBriefs](https://github.com/Ashad001/WebBriefs) (Jul 2024) - [golang-chain](https://github.com/Ashad001/golang-chain) (Jun 2024) - [Climate-Gemma](https://github.com/Ashad001/Climate-Gemma) (Jul 2024) - [Chat-Evaluator-llama3-hackathon](https://github.com/Ashad001/Chat-Evaluator-llama3-hackathon) (Jul 2024) - [PixelSplash](https://github.com/Ashad001/PixelSplash) - [BeyondTheFrame](https://github.com/Ashad001/BeyondTheFrame) (Mar 2025) - [RoomAligner](https://github.com/Ashad001/RoomAligner) (Oct 2024) - [PPT-Automation](https://github.com/Ashad001/PPT-Automation---OpenCV) (Jan 2023) - [2Q48](https://github.com/Ashad001/2Q48) (Feb 2025) - [ML-DL-Projects](https://github.com/Ashad001/ML-DL-Projects) (Mar 2024) - [Qlassify](https://github.com/Ashad001/Qlassify) (Oct 2024) - [EnvWipe](https://github.com/Ashad001/EnvWipe) (Aug 2024) - [EmailSender](https://github.com/Ashad001/EmailSender) (Apr 2024) - [EmailScraping](https://github.com/Ashad001/EmailScraping) (Mar 2024) - [Certificate-Sender](https://github.com/Ashad001/Certificate-Sender) (Nov 2024) - [SEED-Encryption-Lab](https://github.com/Ashad001/SEED-Encryption-Lab) (Sep 2024) - [Monkey on Steroids](https://github.com/Ashad001/monkey-on-steroids) (Jul 2024) - [Truthinator](https://github.com/Ashad001/Truthinator) (Sep 2024) - [2048-Solver](https://github.com/Ashad001/2048-Solver) (May 2024) - [RapidChat](https://github.com/Ashad001/RapidChat) (Jun 2024) - [Latex-Templates](https://github.com/Ashad001/Latex-Templates) (May 2024) - [IR-Indexing](https://github.com/Ashad001/IR-Indexing) (May 2024) ## Experience - Associate Product Manager → Product Manager, Vyro (ImagineArt, Chatly) · Jun 2026 - Present · Islamabad - Founder & Lead Developer, Fyltr · Jul 2025 - May 2026 · Remote - Senior AI Engineer, TraderWare (TradersGPT | Aiversity) · Jul 2025 - Jun 2026 · Karachi - AI Engineer, FireBird Technologies · Aug 2024 - Jun 2025 · Singapore (remote) - Student Teaching Assistant, FAST-NUCES · Sep 2024 - Dec 2024 · Karachi - Junior AI Engineer, RAIN · Sep 2023 - May 2024 · Dallas (remote) ## FAQ ### Who is Ashad Abdullah? Ashad Abdullah is a product engineer based in Islamabad, Pakistan who builds AI products, agents and plugins. He is a Product Manager at Vyro (ImagineArt, Chatly) who stays hands-on: he writes the PRDs and ships the code himself, at the speed of thought, running multiple AI coding agents across parallel workspaces, powered by ADHD. There he shipped ImagineArt into Premiere Pro, After Effects, Photoshop, Figma and Framer, and led an agentic video editor to first release. ### What does Ashad build? AI products and agents for founders and teams, from idea to launch: plugins and extensions that put AI inside existing tools, multi-agent LLM systems, and full products built from zero. Examples include ImagineArt's plugins, the 120+ tool TradersGPT agent and Fyltr. ### Is Ashad available for freelance or full-time work? Yes. He is taking freelance builds and is open to full-time founding engineer, product engineer and builder-PM roles at early-stage AI startups. ### Is Ashad a fit for a founding engineer role? He owns work end to end. He founded Fyltr and took it from idea to launch solo, shipped ImagineArt's plugins into five host apps from spec to release, and led an agentic video editor from prototype to first release in about 8 weeks. He learns whatever stack the problem needs, and works AI-native: multiple coding agents running in parallel workspaces, with him owning what ships. ### Where is Ashad based, and what time zone does he work in? Islamabad, Pakistan, on Pakistan Standard Time (UTC+5). He has worked remotely with teams in Singapore and Dallas. ### Does Ashad work in a specific stack? No. He adapts to whatever the problem needs, and his projects range from assembly to Rust. He picks the tools after the problem, not before. ### How do I start a project with Ashad? Send a brief with the form on this page, saying what you're building, who it's for and when you need it, or email ashad001sp@gmail.com. ## Links - [GitHub](https://github.com/Ashad001) - [LinkedIn](https://linkedin.com/in/ashadqureshi1/) - [X](https://x.com/ashadqu7) - [Resume](https://drive.google.com/file/d/1pSu_AaO_lst8opMBwXwxNhCLYrAmEFPc/view) - [Sitemap](https://ashad001-portfolio.vercel.app/sitemap.xml) - [llms.txt](https://ashad001-portfolio.vercel.app/llms.txt) - [llms-full.txt](https://ashad001-portfolio.vercel.app/llms-full.txt): every case study in full # Case studies in full ## ImagineArt Putting a generation engine everywhere the work already happens Associate PM → PM, surface strategy owner · ImagineArt (Vyro) · Jun 2026 — Present · Shipped, in production Case study: https://ashad001-portfolio.vercel.app/work/imagineart ### What I own ImagineArt reaches millions of users through its own apps. Professional creatives were not among them, because professionals do not leave the software they bill hours in to go generate an asset somewhere else and carry it back. The destination model quietly excludes them. So the question I own is not "how do we get creatives to our product" but "how do we put the product where the work already happens" — which turns out to mean host apps, browsers, and agent runtimes, not one more web page. ### Plugins Five production surfaces — Premiere Pro, After Effects, Photoshop, Figma, and Framer — each a real host runtime with its own sandbox, lifecycle, and distribution story rather than a webview with a logo on it. The Premiere Pro one runs on UXP with a pull-first architecture and ships as a notarized .pkg for macOS. Generate, upscale, clean up shots, and conform Film Studio projects into editable timelines without leaving the edit. ### The browser extension Released on the Chrome Web Store: AI face swap, recreate, and style grab on any page you browse, in a side panel wired to your ImagineArt account. The pitch is the whole thesis in one line — the browser is the world's biggest mood board, so act on any image on the web without leaving the tab. ### Agents and the MCP server Alongside the surfaces I built an agentic video editor, and I maintain ImagineArt's MCP server — moving it toward a lean core tool surface with skills layered on top, so tool selection stops degrading as context grows. That is the same problem I hit at 120+ tools on TradersGPT, arriving from the other direction. ### What the runtimes teach you None of it is documented. What the sandbox will not permit, which integrations look obvious and turn out to be impossible, which constraints only surface in week three once you have already committed to an approach — you learn it by shipping and getting rejected. That accumulated knowledge is the actual asset. It is not a model problem, so the labs will not solve it, and it does not demo well, so most teams will not touch it. ### What happened These surfaces are now a distribution channel into professional users the company previously could not reach. That is company revenue rather than mine, and I would rather say so plainly than dress it up. ## TradersGPT 120+ tools, and a routing layer that decides between them Senior AI Engineer — architecture owner · TraderWare · Jul 2025 — Jun 2026 · In production Case study: https://ashad001-portfolio.vercel.app/work/tradersgpt ### The problem Traders need an answer that pulls from web search, financial document retrieval, and live market data from external APIs — and they need it to be right often enough to rely on. Our agents ran sequentially and took over 200 seconds to fetch, analyze, and visualize. The team treated that as the cost of doing the work properly. ### The opinion I acted on I thought three minutes was a product failure rather than a performance one. Nobody waits three minutes for an answer they can get worse and faster elsewhere. I rewrote the architecture overnight and pitched it at Monday standup: parallel agents, query routing, contextual visualization, and SSE streaming from the backend. Latency dropped to around 30 seconds including thinking time, and because results streamed as they arrived it felt instantaneous. It has been in production since. ### The hard part Tool selection at 120+ tools. Scale breaks routing before it breaks anything else, so most of the design effort went into the layer that decides what to call, not the tools themselves. I built the orchestration, the retrieval, the visualization layer, and the evaluation harness — LangSmith tracing across agent handoffs, tool calls, and orchestration paths — because without it you cannot tell whether a change made the system worse. ### What happened It held uptime that traders were willing to rely on, which is a higher bar than a benchmark. I designed the architecture and implemented it end to end. ## Fyltr Founded it, launched it, and failed to charge for it Founder — solo engineer, solo marketer · Independent · Jul 2025 — May 2026 · Paused Case study: https://ashad001-portfolio.vercel.app/work/fyltr ### What I built An AI-native form builder, founded and built alone. Document extraction, NLP visualization, and Forge — which turned a sentence into a working form. I ran the marketing solo too, through a Remotion pipeline: programmatic React video rendering that generated and published content without manual editing. ### Getting into a room I wasn't invited to The subreddits where my users lived ban self-promotion outright. I had no budget and no obvious way in. So I didn't market the product. I wrote about how I built the launch video in Remotion — which those communities do want — and the product came along with it. 40K+ visitors in launch week, entirely organic, on a post that pulled 60K+ impressions and 2.3K signups. ### Then it stalled Bounce rate sat at 78% and I had not instrumented custom events, so I spent two months debugging a funnel I could not see. The real lesson wasn't the analytics. Form builders are a horizontal category with incumbents who are fine, and I never picked a vertical or found a design partner who would have told me that in week two. ## Bolt-SQL Text-to-SQL that answers before you look away Author · Open source · 2025 · Shipped Case study: https://ashad001-portfolio.vercel.app/work/bolt-sql ### The problem Text-to-SQL is a solved capability and an unsolved experience. The tools work; they just cost you a context switch on every question, which means you stop asking the small questions. ### What I built A text-to-SQL CLI in Rust that ships as a single binary. No runtime to install, no environment to set up, nothing between typing the question and reading the answer. It is the same instinct behind everything else here: I had the problem, the existing answer was slow, so I built the fast one for myself first. ## Motif An MCP server that watches your screen recording and finds the bug Author · Open source · 2026 · Shipped Case study: https://ashad001-portfolio.vercel.app/work/motif ### The problem UI bugs are easy to record and painful to describe. Every reporting flow asks you to translate a video you already have into prose that loses the detail that mattered. ### What I built An MCP server that diagnoses UI bugs directly from screen recordings using Gemini 2.5 Flash — so the coding assistant you are already talking to can watch the GIF itself instead of asking you to describe it. Shipping it as MCP rather than an app is the point. It is the integration-layer bet in miniature: the capability shows up inside the tool you already use instead of asking you to visit one more. ## Devhire A behavioral interview chatbot, built before that was an obvious product Deputy Team Lead · Developers' Day, ACM NUCES · Mar 2023 — May 2023 · Shipped Case study: https://ashad001-portfolio.vercel.app/work/devhire ### What I built An AI-powered behavioral interview chatbot for fresh graduates and interns, co-led at ACM NUCES. OpenAI API against a Flask backend, with TextBlob doing sentiment analysis on the answers. ### Why it still matters to me It is the earliest example of the pattern I have repeated since: notice a capability the moment it becomes real, point it at a problem you personally have, and ship before the category exists. I started as an automation engineer right as the LLM wave hit. This is the project where that turn happened.