Less waiting. More conversation.
For SpeakUp, I replaced separate text and audio transports with one WebSocket stream. A smaller gap between asking a question and hearing an answer.
15s → 2sExplore the case study Open to remote opportunitiesBASED IN PAKISTAN / BUILDING EVERYWHERE
Hamza Asif · Full-Stack AI Engineer
Real-world AI.Built to respond.
THOUGHTFULLY BUILT. READY FOR THE REAL WORLD. Move your cursor to explore
I'm Hamza, a full-stack AI engineer building voice agents, intelligent pipelines, and products people can use. Based in Pakistan. Ready to build with your team, anywhere.
The space between input & impact VOICE / SYSTEMS / AI
VOICE AI · FULL-STACK · APPLIED AIDiscover the work
A few things I've builtSELECTED / 01—03
Currently building voice & outreach agents at ZicsolScroll to the work ↓
A little about how I build
I turn complex technology into things that feel simple. Faster conversations. Smarter workflows. Ideas people can actually use.
A few engineering stories
Small details.
Meaningful differences.
From the first audio chunk to the final experience. Three ways I connect AI to the real world.
Selected work / 01
Built for people.
Engineered for impact.
Voice, language, vision, and 3D.
A closer look at the problems I solve.
WORKFLOW OVERVIEW
01Flutter app
02WebSocket stream
03Text + audio
SpeakUp · freelance project15s → 2s
VOICE AI / FREELANCE/01
A conversation. Without the wait.
SpeakUp — a mobile voice agent with a streaming pipeline rebuilt around a single WebSocket connection.
- Problem
- Text used SSE while audio used request/response. Receiving the first audio chunk took 15 seconds.
- Approach
- Replaced the split transport with one WebSocket stream for both text and audio, connecting the Flutter app to the Node.js backend.
- Result
- Time to the first received audio chunk fell from 15 seconds to 2 seconds.
WORKFLOW OVERVIEW
01English video
02Speech + translation
03Urdu dubbing
Uplift AI / LUMS hackathonEnglish → Urdu
LANGUAGE AI / HACKATHON/02
Awaaz AI
English-to-Urdu video dubbing, built for the Uplift AI / LUMS hackathon.
- Problem
- Make English video content accessible to Urdu-speaking audiences.
- Approach
- Built a seven-step dubbing pipeline using speech processing, translation, Urdu speech generation, and video assembly.
- Result
- An English-to-Urdu dubbing workflow integrating Uplift AI speech technology.
WORKFLOW OVERVIEW
01GLB model
02Real-time conversion
03USDZ / Apple AR
BlenSpark · 3D & ARGLB → USDZ
3D / BLENSPARK/03
From a model to your space.
Real-time GLB-to-USDZ conversion for Apple AR, with AR video optimization.
- Problem
- Bring GLB models into Apple AR workflows that require USDZ.
- Approach
- Implemented real-time model conversion and optimized video for the AR experience.
- Result
- A conversion workflow connecting 3D assets to Apple AR.
WORKFLOW OVERVIEW
01Pakistani telephony
02Python SIP integration
03Voice agent
BlenSpark · voice platform~8s → <2s
VOICE SYSTEMS / BLENSPARK/04
Voice AI, connected to real calls.
Voice agent engineering with a custom Python SIP integration for Pakistani telephony.
- Problem
- Slow voice responses and the need to connect AI conversations to local telephony.
- Approach
- Optimized the voice pipeline and built a custom Python SIP integration.
- Result
- Voice response latency reduced from approximately 8 seconds to under 2 seconds, with approximately 80% lower call costs.
WORKFLOW OVERVIEW
01Room image
02Floor segmentation
03Floor region
BlenSpark · computer visionImage → floor mask
COMPUTER VISION / BLENSPARK/05
See a different floor.
Computer vision floor segmentation for room imagery.
- Problem
- Identify the floor region in a room image for a flooring visualization workflow.
- Approach
- Built a computer vision segmentation project to isolate the floor region.
- Result
- A floor segmentation workflow for visualizing changes in a room.
The person behind the work / 02
Hi, I'm Hamza.
I connect the dots.
~2.5 YEARS
Across full-stack development
and applied AI engineering.
From a mobile voice conversation to a model in augmented reality — I enjoy making complex systems feel simple to use.
I'm a Full-Stack AI Engineer based in Pakistan, currently working at Zicsol as an AI Automation Engineer on voice and outreach agents.
Before that, I grew from an intern into a Full-Stack Developer at BlenSpark, working across voice AI, computer vision, 3D, and web applications. I completed my BS in Artificial Intelligence at UMT in July 2026.
I'm open to international remote roles, freelance collaborations, and open-source work.
See what I'm buildingExperience / 03
A little more
context.
Aug 2026 — PresentCURRENT
AI Automation Engineer
Zicsol
Building voice and outreach agents, connecting AI capabilities to business workflows.
Sep 2024 — Jul 2026
Intern → Full-Stack Developer
BlenSpark
Worked on voice AI, Python SIP integration, floor segmentation, 3D and AR workflows, and full-stack products.
Completed Jul 2026
BS Artificial Intelligence
University of Management and Technology
A foundation in artificial intelligence, complemented by hands-on product engineering.
Built with
The right tools.
For the actual problem.
Frontend & mobile
Backend & data
AI & integrations
Let's make something useful / 04
Your next idea.
Let's build it.
Open to remote full-time roles and freelance projects. If you're building with AI, need a full-stack engineer, or want to collaborate on open source, let's talk.
Start a conversation.
Tell me about your team, the problem you're solving, or a product you want to bring to life.
Connect on LinkedIn