MUDASSAR
AI / ML ENGINEER
AI/ML engineer who designs, trains, and ships production models across deep learning, NLP, computer vision, and large language models. I focus on automation, real-time voice agents, and LLM training — deployed as REST APIs on AWS & Azure.
PLAYER ONE
I build production AI end to end — from voice agents and computer-vision pipelines to large language models.
AI/ML engineer who designs, trains, and ships production models across deep learning, NLP, computer vision, and large language models. I focus on automation, real-time voice agents, and LLM training — deployed as REST APIs on AWS & Azure. I contributed to Qalb, the world's largest Urdu LLM, and I'm currently building across three teams at once.
LEVELS CLEARED
AGENTS LIMITED
Product Owner — CounterVision
Remote
In-store retail intelligence — turning anonymous foot traffic into visitor identity, movement, dwell & conversion analytics from existing cameras.
- ▸Own the product + lead CounterVision v2: real-time person tracking & face recognition, benchmarking 10 model+tracker configs on an RTX 5060.
- ▸Architected the multi-stage pipeline: YOLO26 detection → ByteTrack / BoT-SORT tracking → SCRFD face detection → ArcFace 512-d embeddings. 96.4% tracking continuity at 18.7 FPS.
- ▸Shipped quality-gated recognition (~60-70% compute saved), multi-frame embedding fusion, clothing-color HSV histograms for body re-ID, and self-correction logic.
EMBER
Software Engineer
Remote
Building on Ember's AI knowledge & operator-brain platform. First mission: a full meeting-connector integration, shipped end to end.
- ▸Designed and shipped a new meeting-recorder connector across the whole stack — backend connector, admin-panel UI, and the operator-brain auto-trigger pipeline.
- ▸Indexing → semantic search → cited answers, validated end to end and merged to main with full CI + adversarial review.
NavAI
Voice AI Engineer
Tashkent, Uzbekistan · Remote
Building Uzbek voice agents end to end — custom STT/TTS, real-time flows, observability, and a model playground.
- ▸Built the end-to-end Uzbek voice-agent pipeline: custom STT/ASR + TTS, GPT-4o, LiveKit real-time transport, and SIP telephony (migrated off Yandex).
- ▸Stood up a full observability platform — Prometheus + Grafana + cAdvisor + Node Exporter on GCP, Langfuse tracing, and Telegram alerting via a webhook relay.
- ▸Created a model playground + A/B testing harness and a Post-Call QA agent (MongoDB) detecting language switches, transfers, dead-ends & self-resolution.
- ▸Led the NAV-175 root-cause investigation across 750+ calls — found the non-interruptible greeting driving 22% single-turn calls.
CARECLOUD MTBC
Junior AI Engineer
Rawalpindi, Pakistan
Healthcare AI — claim-denial prediction, medical-coding automation, and a voice agent for doctors.
- ▸Built a claim-denial classifier on millions of 837-EDI insurance claims: 99%+ accuracy on accepted, 80%+ on denied.
- ▸Shipped a CPT/ICD mapping RAG pipeline + clinical-document classification, replacing manual workflows for 500+ staff.
- ▸Built an AI dashboard over 2007–2024 clinical records and contributed to 'Status AI', a voice agent for doctors.
GENESIS LAB
AI Research Intern
Remote
Research on LLMs, GANs, and RAG systems for document automation.
- ▸Researched and implemented LLMs, GANs, and RAG systems for document-automation workflows.
IMARAT GROUP
ML Intern
Islamabad, Pakistan
Classification/regression for real-estate valuation + YOLOv8 detection.
- ▸Trained classification/regression models for real-estate valuation and optimized YOLOv8 detection pipelines.
CARTRIDGES
VANOS
A real-time voice AI desktop assistant with sub-second latency running on consumer hardware.
- +Sub-second end-to-end voice latency on a laptop
- +5 Voice-to-Action (V2A) tools wired into the OS
- +Solved critical mic-compatibility & audio challenges
CLAIM DENIAL
A full ML pipeline for 837-EDI ingestion, classification, and live claim validation with auto-retraining.
- +Millions of insurance claims, 99%+ / 80%+ accuracy
- +Live validation API + automated retraining loop
- +Dockerized & deployed on AWS
DASHGRAB
A computer-vision system that detects items shoppers pick up and auto-generates their bill.
- +Pick-detection from overhead camera streams
- +Pose + segmentation fusion for hand-item association
- +Auto-billing without checkout scanning
PORTAL AGENT
An autonomous agent that automates university-portal tasks end to end.
- +Tool-using LangChain agent over a web portal
- +Plans + executes multi-step portal workflows
- +FastAPI backend with a JS front-end
STAT SHEET
COMPUTER VISION
LV 95VOICE & SPEECH AI
LV 90LLMs & NLP
LV 92ML FRAMEWORKS
LV 93DEPLOYMENT & MLOps
LV 88OBSERVABILITY
LV 85CONTINUE?
I'm open to collaborating, contributing, and building ambitious AI. Drop a coin and let's talk.