Abdul Naafay

AI-Focused Software Engineer & Full-Stack Developer specializing in Agentic Workflows and RAG Pipelines.

Based in: Lahore, PakistanCurrently: Associate Consultant - Management Trainee Officer (MTO), Systems LimitedExperience: 1 year

Hire me

Abdul Naafay is an AI-focused software engineer with extensive experience building production-grade full-stack systems and multi-agent LLM architectures. He bridges the gap between backend engineering and applied AI, having successfully deployed scalable APIs, generative pipelines, and RAG-powered semantic search layers.

Experience

Associate Consultant - Management Trainee Officer (MTO)

Systems LimitedLahore, PakistanworkAug 2026 – Present
Rotating through consulting and technology-delivery engagements, building cross-functional exposure to client-facing enterprise systems, including financial-services accounts.

Trainee AI Engineer

Virtual Force Inc.Lahore, PakistaninternshipJul 2026 – Aug 2026

Summary

Architected a centralized in-house profile management system on a modular backend, retiring legacy spreadsheet workflows across onboarding and role assignment. Engineered a RAG-powered semantic search layer with vector embeddings over an employee-profile directory, enabling natural-language retrieval across unstructured HR records. Sustained 99.5% uptime across 5+ in-house applications via containerized deployment and CI/CD.

What I did

  • Built an internal profile management system to replace spreadsheets for HR and operations onboarding and role assignment.
  • Implemented semantic search functionality to match employee skills with project requirements.
RAGVector EmbeddingsDockerCI/CDSemantic Search

AI & Data Analytics Intern

14x SolutionsLahore, PakistaninternshipSep 2025 – 2026

Summary

Built backend AI-powered Python pipelines for healthcare-records processing, integrating LLM/NLP classification with asynchronous batch processing for 500k+ records monthly. Improved operational efficiency by 60% and cut manual workload by 50% through workflow automation and backend pipeline redesign.

What I did

  • Flagged low-confidence cases for manual review within a pipeline processing 500k records monthly.

Results

  • Processed 500k+ records monthly.
PythonLLMNLPAsynchronous Processing

Skills

Technical

Prompt Engineering
LangChain
RAG Pipelines
OpenAI API
Multi-Agent Orchestration
REST APIs
PythonPython
GitGit
Express.jsExpress.js
JavaScriptJavaScript
SQLSQL
Context & Memory Management
LLM Fine-Tuning (LoRA, DPO)
LangGraph
Hugging Face
FastAPIFastAPI
Node.jsNode.js
MongoDBMongoDB
PostgreSQLPostgreSQL
RedisRedis
DockerDocker
JWT Auth
Swagger/OpenAPI
SQL ServerSQL Server
Pinecone
FAISS
WebSockets
System Design
TypeScriptTypeScript
CI/CD
Human-in-the-Loop Design
pgvector
RBAC
LoRA Fine-Tuning
DPO Alignment
Computer Vision
Semantic Search

Projects

Isekai - Text-Based RPG with Sentient AI Agents

Lead Developerpersonal
Designed a multi-agent architecture powering persistent, sentient NPCs with long-term memory and emergent, context-aware dialogue orchestrated via LangChain over LLM backends. Built a dynamic narrative engine with procedurally generated storylines and player-driven state changes (+45% avg. session length); agent memory/context-window management cut repetitive and hallucinated responses by 35%.
PythonLangChainLLM AgentsFastAPIWebSockets

Prompt-to-Video AI Pipeline

Developerpersonal
Built an end-to-end generative pipeline converting an image + text prompt into a talking-head video with synced audio using SadTalker, TTS, and diffusion-based lip-sync. Optimized inference via batching, caching, and GPU tuning, cutting generation time from 8 to 3 minutes/video (62% reduction).
PythonSadTalkerDiffusion ModelsTTSFFmpeg

IQueue

Full Stack Developerpersonal
Deployed a full-stack digital queue-management app serving 100+ users via 30+ REST endpoints, achieving sub-350ms latency with Redis caching, Docker, and CI/CD; improved throughput by 25%.
ReactNode.jsExpressMongoDBRedisDockerCI/CD
Visit the site

SmolLM: RoPE, GQA & Preference Alignment (LoRA/DPO)

AI Researcher/Developerpersonal
Implemented a Llama3-style decoder (RMSNorm, RoPE, Grouped-Query Attention, SwiGLU) from scratch, then applied LoRA for parameter-efficient fine-tuning and Direct Preference Optimization (DPO) to align outputs with human preference data.
PyTorchTransformersLangChainLoRADPO

3D Scene Reconstruction & Virtual Tour

Developeracademic
Built an incremental SfM pipeline (feature matching, pose estimation, bundle adjustment) and a browser-based Three.js viewer with pose-interpolated, photorealistic virtual walkthroughs.
PythonStructure-from-MotionThree.js

Edulink

Full Stack Developerpersonal
Built a full-stack learning platform with 20+ APIs and real-time messaging, improving onboarding efficiency by 50% via scalable schema design.
ReactNode.jsExpressMongoDBWebSocketsReal-time messaging

Education

B.S. in Computer Science

Lahore University of Management Sciences (LUMS)Artificial IntelligenceGPA 3.56Aug 2022 – Jul 2026

Coursework

Artificial IntelligenceMachine LearningGenerative AINLPComputer VisionData MiningDatabase SystemsOperating SystemsSoftware EngineeringNetwork Security

Societies & activities

Teaching Assistant - Advanced ProgrammingTeaching Assistant - Database Systems

Recognition

Awards

Dean's Honour List

Lahore University of Management Sciences (LUMS)2025

Awarded for Fall '23, Spring '24, and Spring '25 semesters.

Hire AbdulReach out about a role, a contract or a conversation.For recruiters

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

More people on Kanvis

Browse more peopleCreate your own Kanvis