I’m Salman, a data scientist and AI engineer based in Islamabad, Pakistan, focusing on building production-ready AI systems that scale in media and entertainment. I enjoy turning complex research into practical tools—LLM-powered RAG, vision pipelines, and multi-agent metadata mapping—delivering fast, reliable solutions across FastAPI, Docker, and cloud deployments.…

Salman Asad

I’m Salman, a data scientist and AI engineer based in Islamabad, Pakistan, focusing on building production-ready AI systems that scale in media and entertainment. I enjoy turning complex research into practical tools—LLM-powered RAG, vision pipelines, and multi-agent metadata mapping—delivering fast, reliable solutions across FastAPI, Docker, and cloud deployments.…

Available to hire

I’m Salman, a data scientist and AI engineer based in Islamabad, Pakistan, focusing on building production-ready AI systems that scale in media and entertainment.

I enjoy turning complex research into practical tools—LLM-powered RAG, vision pipelines, and multi-agent metadata mapping—delivering fast, reliable solutions across FastAPI, Docker, and cloud deployments.

See more

Language

English
Fluent

Work Experience

Data Scientist at Upright Music/CEBS
July 1, 2024 - Present
Evaluated and deployed DeepSeek-R1 LLM in production, comparing performance across VLLM, Ollama, Llama.cpp, and SGLang; selected Llama.cpp for production, optimizing GPU memory efficiency and inference speed. Designed and optimized an LLM-powered agentic RAG system for natural-language music search, deployed as a FastAPI endpoint. Improved Triton Inference Server configuration to reduce GPU usage by 30–40%, enabling multi-model deployment on shared infrastructure. Built a multi-agent system for mapping music metadata tags between taxonomies using LangGraph, FAISS, DeepSeek-R1, and prompt-controlled iterative routing with retry logic.
Computer Vision Engineer at xis.ai
June 14, 2021 - July 1, 2024
- Trained and optimized models to ONNX and TensorRT formats, improving inference speed for real-time pipelines. - Conducted research on anomaly-detection frameworks, SAHI Inferencing and Anomalib, for visual anomaly detection and quality inspection. - Built high-performance detection systems (YOLOv10, RT-DETR) with focus on reliability and throughput— experience transferable to benchmarking LLM performance.
Computer Vision Engineer at Strada Imaging
March 16, 2020 - July 12, 2021
• Prepared and validated large-scale road imagery datasets across potholes, traffic lights, traffic poles, lane markings, pavement patches, and other infrastructure classes for detection and segmentation models. • Implemented an OpenCV road-patch detection pipeline using Rosin’s unimodal thresholding, closed-contour analysis, and texture similarity based on a published academic methodology.

Education

Bachelor of Science in Computer Science at Air University
September 15, 2018 - June 24, 2022

Qualifications

Add your qualifications or awards here.

Industry Experience

Media & Entertainment, Software & Internet
    AI-Powered Support and Operations Dashboard

    Next.js, PostgreSQL, FastAPI, LangChain, OpenAI API
    • Owned delivery of a full-stack internal operations dashboard connecting a Next.js frontend, FastAPI backend, PostgreSQL, authentication, role-based access control, and LLM-powered workflows.
    • Worked with the client to translate operational requirements into user flows, API contracts, access-control rules, assistant behavior, and delivery milestones.
    • Integrated an LLM assistant with schema-constrained responses, prompt guardrails, validation, fallback handling, and error recovery, then packaged the stack with Docker and environment-based configuration.

    Contract Intelligence Agent

    • Built a full-stack application that ingests contracts, extracts structured clauses and obligations, identifies potential risks, and supports document-grounded question answering with source references.
    • Designed modular FastAPI services for document processing, retrieval, agent orchestration, and schemaconstrained outputs, connected to a Next.js frontend and PostgreSQL persistence layer.
    • Added Docker-based local deployment, environment configuration, automated tests, and CI checks to keep frontend, backend, and document-processing changes reproducible.