m
mqasimali194

M Qasim Ali

@mqasimali194

AI Engineer

Pakistan
Englisch, Urdu
Einige Informationen werden in englischer Sprache angezeigt.
Über mich
I’m Muhammad Qasim Ali, an AI Engineer specializing in enterprise AI solutions. I build solutions across Agentic AI, LLMs, RAG, Computer Vision, Document Intelligence, and Voice AI. My expertise includes custom AI agents, workflow automation, hybrid retrieval, OCR, intelligent document processing, and self hosted LLM deployments. Tech Stack: Python, PyTorch, FastAPI, LangChain, LiveKit, vLLM, Docker, and PostgreSQL. I focus on scalable, reliable AI systems that solve real business problems and move beyond prototypes into production.... Mehr lesen

Kompetenzen

m
mqasimali194
M Qasim Ali
offline • 

Meine Dienstleistungen

KI-Integrationen
I will build, train, and deploy neural networks for deep learning projects
Entwicklung von KI-Chatbots
I will build rag chatbots with custom knowledge integration

Portfolio

Arbeitserfahrung

TMC

Sovereign AI Engineer

TMC

Jun 2026 - Present • 5 mos

• Architected and delivered enterprise AI solutions spanning LLMs, computer vision, RAG, fraud detection, agentic workflows, and real-time voice systems. • Led the development of backend services and integration of self hosted LLM, ASR, and TTS models, leveraging SGLang, LiveKit, and Pipecat to build low latency conversational AI systems. • Designed and optimized on premises AI infrastructure and inference architectures, addressing scalability, performance, reliability, data privacy, and production deployment requirements.

PITB_Incubation Wing

AI Engineer

PITB Incubation Wing

Apr 2025 - Sep 2025 • 5 mos

● Engineered a U-Net semantic segmentation pipeline predicting soil moisture index masks from 13-band Sentinel-2 multispectral imagery (10–60 m resolution), achieving 0.89 mean IoU, with MLflow for experiment tracking, model versioning, and performance monitoring. ● Deployed a hybrid search RAG chatbot with hallucination guardrails and reranking, improving retrieval accuracy by 22% and cutting irrelevant responses by 30%