Associate AI Engineer • Mayfair Asian Food Industries

Islam Nabi

AI/ML Engineer · Agentic AI Developer

Building Production-Grade AI Systems

  • Agentic AI & Generative AI
  • RAG & LLM Systems
  • Local LLM Deployment
  • Speech AI
  • Computer Vision
  • Machine Learning & Deep Learning
  • MLOps
  • Python

Every aspect of learning or any other feature of intelligence can, in principle,
be so precisely described that a machine can be made to simulate it.

— John McCarthy, 1956

About Me

I'm an AI/ML Engineer based in Lahore, Pakistan, building production-grade AI Systems. My focus is on Agentic AI, Generative AI, RAG and LLM Systems, Speech AI, Computer Vision, Machine Learning & Deep Learning, MLOps and Python, from architecture to deployment and Monitoring.

I've trained LLMs, built multi-agent systems, Personal Assistants, Intelligent RAG and LLM-Powered Systems and delivered real-time CV pipelines for enterprise clients. I hold a BS in Computer Science from Sukkur IBA University and currently working as an Associate AI Engineer at Mayfair Asian Food Industries.

"Engineering intelligence that empowers humanity."
Current RoleAssociate AI Engineer @ Mayfair
EducationBS CS, Sukkur IBA University
LocationLahore, Pakistan
Islam Nabi | AI/ML Engineer

Experience Timeline

  1. Associate AI Engineer — Mayfair Asian Food Industries Ltd.

    Lahore ·

    Building and deploying production-grade Agentic AI, Gen AI, CV, RAG and LLM-Powered Systems, specializing in autonomous workflows, multi-agent architectures, and intelligent enterprise applications.

  2. AI Trainer — Code & Reasoning Evaluator — Revelo

    Remote ·

    Evaluated and improved AI coding and reasoning systems by assessing code generation, multi-turn reasoning, and AI-generated solutions across real-world open-source projects. Leveraged LLM evaluations, RLHF, and structured feedback frameworks to enhance model performance, reliability, and reasoning capabilities.

  3. AI/ML Engineer Intern — PureLogics Software House

    Lahore ·

    Developed end-to-end Urdu Speech AI System, building large-scale audio dataset preparation and transcription pipelines, Speech Recognition (ASR), Text-to-Speech (TTS), speech enhancement, and voice cloning solutions. Leveraged Whisper, Silero VAD, FastAPI, LoRA fine-tuning, and MLOps practices to create scalable, production-ready pipelines and curate over 100 hours of high-quality Urdu speech data for model training and research.

  4. AI/ML Engineer Trainee — PureLogics Software House

    Lahore ·

    Completed an intensive AI/ML training program covering Python, data preprocessing, Machine Learning, Deep Learning, Generative AI, RAG systems, and AI Agents. Applied NumPy, Pandas, and Scikit-learn for data cleaning, transformation, feature engineering, and model development, while building end-to-end intelligent applications using TensorFlow, PyTorch, LangChain, LangGraph, and LLMs, ending in the development and deployment of the SmartInsureAI capstone project.

  5. Software Engineer Intern — InternnCraft

    Remote ·

    Developed full-stack web applications using HTML, CSS, JavaScript, and React, building responsive user interfaces and integrating backend functionality through APIs, databases, and dynamic data handling. Contributed to reusable component development, database integration, application debugging, and performance optimization to deliver scalable and maintainable software solutions.

  6. LLM Trainer — Remotasks (San Francisco, USA)

    Remote ·

    Trained, evaluated, and optimized Large Language Models (LLMs) for code generation and reasoning tasks by analyzing AI-generated outputs, comparing solutions, and providing structured feedback to improve model accuracy, reasoning capabilities, and response quality through human-in-the-loop evaluation.

Featured Projects

Production-grade AI systems designed for real-world impact and scalability.

Agentic AI

AI Sales Intelligence System

Agentic AILangGraphReAct AgentsText-to-SQL AgentsFastAPIPostgreSQLClaudeNext.jsAWS
  • Built a LangGraph-powered AI Sales Agent enabling natural language analytics over a 50M+ record PostgreSQL sales warehouse.
  • Engineered a ReAct-based autonomous agent that dynamically plans, executes SQL queries, and synthesizes management-ready insights.
  • Developed hybrid intelligence by combining internal sales data with live web search for competitor benchmarking and market analysis.
  • Designed a persistent memory architecture that stores business events, entities, and conversational context across sessions.
  • Implemented real-time streaming APIs with FastAPI and SSE, powering interactive dashboards and conversational analytics in Next.js.
  • Created pre-aggregated KPI pipelines and dynamic chart generation for instant trend analysis, regional insights, and SKU performance monitoring.
Read case study →
RAG

AI-Based HS Code & Landed Cost Estimation

RAGFastAPIFAISSLangChainOpenAI GPT-4o-miniAWS EC2PostgreSQLPDF ProcessingVector SearchREST APIs
  • Built a production-grade RAG system over Pakistan's customs tariff database with 7,500+ indexed HS codes and 93%+ hierarchical category coverage.
  • Engineered an 8-step AI classification pipeline combining FAISS retrieval, GPT-4o-mini reasoning, soft chapter boosting, and multi-factor ranking to achieve ~99% classification accuracy.
  • Developed table-aware PDF extraction and smart chunking pipelines that preserve tariff hierarchies and semantic relationships across FBR documents.
  • Implemented a complete FBR-compliant landed cost engine supporting sequential customs duties, taxes, currency conversion, and TCS shipping calculations.
  • Designed scalable FastAPI services and REST APIs for real-time HS code retrieval and e-commerce-ready landed cost estimation.
  • Deployed the end-to-end system on AWS EC2 with an interactive web interface enabling automated import cost intelligence.
Read case study →
Speech AI

Urdu Text-to-Speech System

Speech AIDataset PreparationText-to-SpeechOpen-Source TTS ModelsWhisperSilero VADLoRAQ-LoRAFastAPIPyTorchAudio ProcessingPython
  • Built an end-to-end Urdu Speech AI System covering data collection, preprocessing, model training, and real-time speech synthesis.
  • Engineered a 100+ hour high-quality Urdu speech corpus using Silero VAD, audio enhancement, and Whisper-based transcription pipelines.
  • Developed automated dataset curation pipelines for speech segmentation, noise reduction, normalization, and metadata generation.
  • Fine-tuned open-source Text-to-Speech models using LoRA/Q-LoRA to generate natural and high-quality Urdu speech.
  • Designed scalable FastAPI-based inference services supporting text normalization and low-latency audio generation in WAV/MP3 formats.
View project →
Computer Vision

Employee Attendance System (CCTV-Based)

Computer VisionFace RecognitionInsightFaceFAISSOpenCVPostgreSQLFastAPINext.jsRTSP StreamingPython
  • Built a real-time multi-camera attendance system using RTSP CCTV streams and AI-powered face recognition with 95%+ identification accuracy.
  • Engineered an end-to-end face recognition pipeline using InsightFace Buffalo, generating 512-dimensional embeddings for robust identity matching.
  • Implemented FAISS-based vector search with cosine similarity for low-latency employee identification across large embedding datasets.
  • Developed an automated attendance engine supporting check-in/check-out logic, duplicate prevention, shift handling, and overtime calculations.
  • Designed a Next.js HR dashboard for employee lifecycle management, live camera monitoring, attendance analytics, and report generation.
  • Built a scalable PostgreSQL-backed architecture supporting 1,000+ employees, 6–10 simultaneous cameras, and real-time attendance processing.
View project →
MLOps

Smart Vehicle Insurance Risk Assessment

MLOpsScikit-learnFastAPIMongoDB AtlasAWS S3AWS EC2AWS ECRDockerGitHub ActionsCI/CDPython
  • Built an end-to-end MLOps pipeline for vehicle insurance risk assessment, covering data ingestion, model training, deployment, and monitoring.
  • Engineered automated data pipelines for validation, transformation, feature engineering, and model evaluation using Scikit-learn.
  • Developed and deployed real-time prediction APIs with FastAPI for low-latency insurance risk scoring and inference.
  • Implemented model versioning and artifact management using AWS S3 to ensure reproducible and scalable ML deployments.
  • Designed a production-grade CI/CD workflow using GitHub Actions, Docker, Amazon ECR, and EC2 for automated build and deployment.
  • Built a cloud-native architecture with monitoring and logging capabilities for reliable and scalable machine learning operations.
Read case study →
Computer Vision

Wrist Abnormality Detection System

Computer VisionMedical ImagingYOLOv9Transfer LearningFlaskGoogle GeminiSQLitePDF GenerationOpenCVPython
  • Built an AI-powered clinical decision support system for wrist X-ray analysis using a custom fine-tuned YOLOv9 model on the GRAZPEDWRI-DX dataset.
  • Engineered a complete computer vision pipeline including data preprocessing, augmentation, transfer learning, and hyperparameter optimization for abnormality detection.
  • Developed a real-time Flask inference API for secure image upload, automated detection, and annotated X-ray generation.
  • Integrated Google Gemini to generate structured diagnostic reports containing findings, severity assessment, and medical recommendations.
  • Designed a full-stack medical reporting workflow supporting PDF generation, image export, authentication, and session management.
  • Implemented secure data management and logging infrastructure for handling patient information, uploaded images, and AI-generated reports.
View project →
Local LLM

OfflineMail AI: Offline AI Email Assistant

OllamaQwen3FastAPINext.jsTypeScriptSSE StreamingSQLiteLocal LLM Inference
  • Built a fully offline AI email assistant with zero cloud dependency: every generation, rewrite, and tone-change runs on a local LLM via Ollama, with no data ever leaving the device.
  • Engineered a transport-layer privacy enforcement mechanism using a custom httpx transport that allowlists only localhost connections, making the offline guarantee structurally enforced rather than just configured.
  • Designed a real-time SSE streaming pipeline from FastAPI to Next.js so token-by-token generation feels instant despite running an 8B-class model locally.
  • Built five distinct AI-writing features sharing one prompt-orchestration and JSON-output-parsing core, avoiding duplication across features.
  • Implemented structured output enforcement via Ollama's JSON schema mode combined with /no_think prompting to eliminate reasoning-chain leakage from Qwen3 into user-facing output.
  • Delivered a complete local persistence layer (SQLite + async SQLAlchemy + Alembic migrations) for email history, with no external database required.
Read case study →
Full Stack

Visitor Management System

Full-Stack DevelopmentNext.jsReactTypeScriptMongoDBAWS EC2AWS S3REST APIsTailwind CSSRBAC
  • Built a cloud-based Visitor Management System for secure visitor registration, tracking, and analytics with role-based access control.
  • Developed a full-stack application using Next.js and TypeScript supporting visitor registration, vehicle management, blacklist verification, and one-click check-in/check-out.
  • Engineered secure media handling with AWS S3 private buckets and pre-signed URLs for storing visitor photos and CNIC images.
  • Designed scalable backend APIs with server-side pagination, advanced filtering, search, and CSV export for visitor and blacklist management.
  • Implemented real-time dashboards and reporting features for visitor statistics, duration tracking, and administrative analytics.
  • Deployed the entire platform on AWS EC2 with MongoDB Community Edition, including automated database snapshots and recovery mechanisms.
View project →
Full Stack

CX Pricing Calculator: Real-Time Sales Pricing Tool

Next.jsTypeScriptTailwind CSSOpenAI APIApp Router
  • Built an internal real-time pricing calculator enabling customer experience agents to compute accurate product pricing live during customer calls.
  • Integrated OpenAI's GPT-4o-mini to auto-estimate product shipping weight from title and category, with a deterministic heuristic fallback when AI estimation isn't available.
  • Designed a pure, UI-decoupled calculation engine driven entirely by a single configurable rates file, letting business teams update pricing rules without touching application code.
  • Implemented multi-factor cost computation across purchase platform type, delivery/logistics tier, regional zone, and applicable surcharges.
  • Built a real-time summary panel that recalculates the full cost breakdown instantly as agents adjust any input.
  • Delivered multi-currency display (primary/secondary currency conversion) for cross-market pricing clarity.

Technical Expertise

Production capabilities across agentic workflows, speech AI, computer vision, and MLOps, the differentiating layer of my work.

Agentic Workflows

LangChain, LangGraph, ReAct Agents, Tool-Use Agents

RAG Pipelines

FAISS, Vector DBs, Hybrid Search, Contextual Retrieval

LLM Orchestration

LangChain, Prompt Engineering, Function Calling, Chains

Speech AI Systems

Whisper (ASR), VITS, XTTS, VAD, Speaker Embeddings

LLM Fine-Tuning

LoRA, PEFT, Hugging Face Transformers, WER Evaluation

AI Deployment

FastAPI, Streamlit, Flask, Docker, AWS EC2/S3

MLOps & Monitoring

MLflow, CI/CD Pipelines, GPU Inference Optimization

Computer Vision

YOLOv9, OpenCV, CNNs, Transfer Learning, Face Recognition

Tech Stack

Core tools I use to build and ship AI systems.

Languages

Python
JavaScript
TypeScript
C++
Assembly
C
Java
Bash
Rust
Go

AI / ML Core

PyTorch
TensorFlow
Scikit-learn
Hugging Face Transformers
Ollama
Keras
XGBoost
Pandas
NumPy
YOLO
CNNs
OpenCV

Generative AI

RAG
Embeddings
Prompt Engineering
LLM Fine-Tuning
PEFT
NLP
Transformers
FAISS
Flask
Streamlit

Agentic AI

LangChain
LangGraph
ReAct Agents
Multi-Agent Systems
Function Calling
Model Context Protocol (MCP)
REST APIs
Model Deployment
Model Monitoring
ETL Pipelines

Speech AI

Whisper
VITS
XTTS-v2
Silero VAD
Speaker Embeddings
LoRA / QLoRA Fine-Tuning
Dataset Preparation
Open-Source LLMs
FFmpeg
Transfer Learning
InsightFace

Databases

SQL
PostgreSQL
MongoDB
Vector Databases
ChromaDB
Weaviate
Milvus
Polars
Redis
SQLite

Infrastructure & MLOps

Docker
AWS (EC2 / S3 / ECR)
FastAPI
GitHub Actions
MLflow
DVC
Nginx
CI/CD
Gunicorn
Kubernetes

Frontend

HTML
CSS
JavaScript
React JS
Next JS
Tailwind CSS
Sass
npm
Git
Webpack

AI Services

End-to-end AI engineering from strategy and architecture to production deployment. I build reliable, scalable, and business-driven AI systems that deliver measurable real-world impact.

01

Agentic AI Systems

Design and development of autonomous AI systems powered by ReAct and multi-agent architectures. Includes tool-use orchestration, Text-to-SQL agents, workflow automation, and production-grade reasoning systems.

02

RAG & Knowledge Systems

Production retrieval systems over private data and enterprise knowledge bases, featuring hybrid search, vector indexing, contextual retrieval, and domain-specific intelligence pipelines.

03

Speech AI Engineering

Custom speech systems for multilingual and specialized domains, including ASR, TTS, voice cloning, speaker verification, and GPU-optimized transcription pipelines.

04

Computer Vision Systems

Real-time computer vision solutions for surveillance, healthcare, and industrial applications, including facial recognition, medical imaging, and large-scale object detection systems.

05

LLM Fine-Tuning & Evaluation

Task-specific adaptation of foundation models using parameter-efficient training, evaluation pipelines, dataset curation, benchmarking, and production optimization.

06

MLOps & AI Deployment

End-to-end AI infrastructure covering CI/CD pipelines, containerization, cloud deployment, model versioning, observability, and scalable inference services.

Certifications

Industry-recognized certifications and professional training in Artificial Intelligence, Machine Learning, Large Language Models, and Software Engineering.

AI/ML Engineering Internship

PureLogics ·

Hands-on AI/ML engineering across speech recognition, NLP, LLMs, and production MLOps workflows.

Speech RecognitionASRNLPLLMsMLOps
Verify →

Python + AI Bootcamp

PureLogics ·

Intensive Python and AI training covering machine learning, deep learning, and data science fundamentals.

PythonNumPyPandasDeep Learning
Verify →

Supervised Machine Learning: Regression and Classification

Coursera ·

Core supervised learning foundations — regression, classification, and linear algebra from Andrew Ng's program.

Machine LearningRegressionClassification
Verify →

Developing LLM Applications With LangChain

DataCamp ·

Building production LLM applications with LangChain — chains, agents, and real-world deployment patterns.

LangChainLLM ApplicationsNLP
Verify →

Working With Hugging Face

DataCamp ·

Practical NLP with Transformers — model hub, tokenizers, and pipeline APIs via Hugging Face.

TransformersHugging FaceNLP
Verify →

Engineering Insights

Deep dives into the systems I've built: architecture decisions, lessons learned, and results.

Building an offline AI email assistant with Ollama and Qwen3

How I built OfflineMail AI, a privacy-first email writer where every inference runs on-device via local LLMs, with SSE streaming and transport-layer privacy enforcement.

Read case study →

How I built a 95%-accurate RAG system over 7,500 FBR tariff codes

A breakdown of the chunking strategy, retrieval pipeline, and the 8-step classification engine that powers the HS Code Estimation system.

Read case study →

Building a Text-to-SQL ReAct agent that never hallucinates

How I constrained an LLM agent to SELECT-only queries with row limits and retry caps, and why grounding beats prompting alone.

Read case study →

Building a production-grade MLOps pipeline for vehicle insurance risk assessment

How I built SmartInsureAI with FastAPI, MongoDB, AWS, Docker, and GitHub Actions, from data ingestion to live prediction APIs.

Read case study →

Let's Build Something Together

Open to collaborations on Agentic AI systems, LLM applications, and production ML pipelines.