Mujtaba Nasir

AI Engineer specializing in production-grade GenAI and scalable cloud-native solutions.

Based in: Lahore, PakistanCurrently: AI Engineer, PureLogicsExperience: 2 years

Hire me

Mujtaba is an ambitious AI Engineer with extensive experience building RAG pipelines, document intelligence platforms, and multimodal AI systems across diverse industries. He excels at translating complex business requirements into high-performance technical architectures using LangChain, FastAPI, and cloud-native technologies.

Experience

AI Engineer

PureLogicsLahore, PakistanworkMar 2025 – Present

Summary

Developed AI-driven systems across Fitness & Retail Analytics (300+ gym locations), US Government Legislative Archives (80,000+ documents) for the Government of Alaska, NGO Intelligence, Legal tech, and Forensic Investigation domains, delivering scalable cloud-native AI solutions on GCP and AWS. Led client-facing meetings, architecture discussions, requirement gathering, technical initiatives and translating ambiguous business problems into production-grade AI pipelines. Architected and maintained end-to-end Gen AI RAG pipelines integrating multimodal (text + image) knowledge bases and Voice AI workflows (STT/TTS, diarization, transcription, meeting summarization), implementing guardrails, retrieval grounding, hallucination mitigation, and LLM token optimization to improve response accuracy and reduce inference cost. Leveraged LangChain and agentic AI orchestration for structured reasoning, tool routing, and fine-tuning of pretrained NLP and Vision models for domain-specific applications. Engineered a scalable FastAPI + PostgreSQL backend platform powering an AI-driven crawler → scraper → LLM enrichment pipeline with lifecycle orchestration, confidence scoring algorithms, RBAC-secured REST APIs, concurrent batch processing, and Docker-based deployment.

What I did

  • Optimized chunk retrieval and filtered irrelevant context to improve retrieval relevance and reduce hallucinations.
  • Managed LLM token budgets to reduce unnecessary token consumption.

Results

  • Delivered AI-driven systems for 300+ gym locations and 80,000+ government documents.
  • Implemented LLM token optimization to reduce inference cost.
GCPAWSGen AIRAGSTTTTSLangChainAgentic AINLPVision modelsFastAPIPostgreSQLLLMRBACREST APIsDockertiktokensemantic chunkingtop-K retrievalcontext filtering

Junior AI/ML Engineer

Hazen Technologies IncLahore, PakistanworkSep 2024 – Feb 2025

Summary

Developed an end-to-end AI Document Intelligence platform automating Personal Injury Protection (PIP) claim processing across 20+ US insurance providers, reducing manual review time by over 60% and improving operational throughput. Designed and optimized a scalable OCR pipeline using Azure Document Intelligence (Prebuilt Read + Layout) to extract structured data from PDFs, cheques, EOBs, invoices, and medical reports, achieving 91% field-level extraction accuracy across diverse document formats. Built scalable backend APIs using Python and Azure Functions, enabling real-time document ingestion, asynchronous processing workflows, and reliable AI model deployment on Azure Cloud.

What I did

  • Defined Pydantic models for extracting fields such as claim IDs, patient information, service dates, and billing amounts in document intelligence workflows.

Results

  • Improved data reliability by using Pydantic and Instructor to enforce schema consistency and type validation for extracted fields.
  • Reduced manual review time by over 60%.
  • Achieved 91% field-level extraction accuracy across diverse document formats.
Azure Document IntelligenceOCRPythonAzure FunctionsAzure CloudPydanticInstructorStructured Decoding

Skills

Technical

PythonPython
FastAPIFastAPI
Data Science
Machine Learning
Azure Cloud
MySQLMySQL
LangChain
Deep Learning
FlaskFlask
DockerDocker
NLP workflows
4-bit quantization (Q4_K_M)
Natural Language Processing
Data Science Tools and Techniques
Pydantic
Instructor
Ollama
Embedding-based retrieval
Prompt context size optimization
GGUF
Quantization
tiktoken
semantic chunking

Areas of expertise

Analytical
Team Player
Team Work
Communication
Leadership

Languages

English

Projects

YouTube Comment Intelligence System

personal
Developed an end-to-end pipeline for large-scale data collection, preprocessing, text analysis, and insights extraction from noisy, unstructured data.
NLP

Locally hosted RAG chatbot

personal
Built a locally hosted RAG chatbot integrating an open source LLM with document retrieval and context aware responses.
Ollama

Local Inference on Resource-Constrained Hardware

personal
Used GGUF quantized LLMs for local inference to reduce memory consumption while maintaining reasonable inference quality and performance.
GGUFQuantized LLMs

Education

Master of Science (MS)

National University of Computer and Emerging Sciences (FAST-NUCES)Data Science (AI)2026

Bachelor of Science (BS)

University of Central PunjabComputer ScienceGPA 3.91Jul 2024
Hire MujtabaReach out about a role, a contract or a conversation.For recruiters

Mujtaba's twin is AI, it can make mistakes.

More people on Kanvis

Browse more peopleCreate your own Kanvis