Available for opportunities · Karachi, Pakistan

HELLO, I'M

Rahul [ detected: rahul_kumar · 99.7% ] Kumar.

I'm an AI Engineer

Building autonomous agents, computer vision systems, and RAG pipelines. Currently engineering automation at Technocas and training in advanced AI at NAVTTC PMYSDP @ IBA Karachi.

View my work → Download résumé ↓

location · karachi.pk
status · online
role · ai_engineer

01 — Get to know me

About me.

[ rahul_kumar · 99.7% ]

SZABIST Karachi · BS Artificial Intelligence · Class of 2027

I'm a third-year Bachelor of Artificial Intelligence undergraduate at SZABIST Karachi, focused on shipping production AI systems — not just training models in notebooks.

I've built an AR campus navigator that won 1st place at ZabFest 2026 using a fine-tuned DINOv2 ViT-B/14 at 98.3% accuracy. A custom YOLOv8 model that took home the IBA × Duality AI challenge. And a customer-propensity classifier on 11M+ records that placed runner-up at Evolve'25.

Currently engineering automation pipelines at Technocas and training in advanced AI at the NAVTTC PMYSDP residency at IBA Karachi. Elected Program Representative for SZABIST's BS AI department.

Open to collaborations in Agentic AI, Computer Vision, RAG systems, and MLOps.

Hackathon Wins

98.3%

Best Model Acc

11M+

Records Processed

100+

Students Represented

02 — Where I've worked

Experience.

AI Automation Engineer

Technocas · Karachi, Pakistan

Apr 2026 — Present

Engineering automated workflows for an AI services startup — translating business processes into scalable Python agent systems and serverless pipelines.

Architected and deployed an automated CRM system, minimizing human intervention across customer data pipelines.
Engineered a fully automated project-management workflow within Wekan, eliminating manual task tracking.
Designed serverless automation pipelines with LangChain and Zapier to ingest unstructured documents and trigger downstream API actions.
Built agentic AI flows using n8n for cross-platform task orchestration.

PythonLangChainn8nZapierWekan APIAgentic AI

AI Trainee · Fully Funded Residency

NAVTTC PMYSDP · IBA Karachi

Feb 2026 — Present

Selected for an intensive 3-month diploma in Deep Learning and AI Communication, hosted at the Institute of Business Administration under industry mentorship.

Advanced coursework in CNN, RNN, ANN architectures with hands-on PyTorch implementations.
Production-grade MLOps: deployment, monitoring, and lifecycle management of ML systems.
Industry-mentored capstone projects on real-world AI applications.

Deep LearningPyTorchMLOpsCNN / RNN / ANN

03 — Deep dive

Case study: ZabLens.

★ 1ST PLACE · ZABFEST 2026 · ML COMPETITION WINNER

An AR indoor navigator for SZABIST 100 Campus that sees where you are and shows you where to go.

No GPS. No beacons. No QR codes. Just a smartphone camera, a fine-tuned vision transformer, and a graph.

▸ WHAT

A camera-based AR navigator for SZABIST 100 Campus.

Point your phone at any corridor, doorway, or landmark — the system identifies your location from the live frame and overlays a real-time directional arrow guiding you to your destination across 13 indoor nodes.

▸ WHY

Indoor navigation is the gap GPS can't fill.

Google Maps stops working the moment you step inside. New students, visitors, and parents get lost daily in the multi-floor 100 Campus. Static printed maps fail because they require you to already know where you are.

▸ HOW

DINOv2 + FAISS + A* + Gyroscope AR.

A fine-tuned vision transformer recognizes the location from one frame. FAISS finds the closest match. A* computes the shortest path on a campus graph. The gyroscope orients an AR arrow. Re-localizes every 3.5 seconds.

Vision Backbone

DINOv2 ViT-B/14

Test Accuracy

98.3%

Custom Dataset

630 imgs · 13 classes

Retrieval

FAISS IndexFlatIP

Pathfinding

A* on Campus Graph

Companion Chatbot

ZabGPT (RAG · Groq Llama 3)

Camera Frame → DINOv2 768-d Embedding → FAISS Cosine Search → Top-K Majority Vote → A* Path → Gyro-Aligned AR Arrow

◆

Outcome. Re-localizes every 3.5 seconds. Runs on a laptop, accessed from any phone via HTTPS tunnel — no app install, no special hardware. Won 1st place AND the ML Competition at ZabFest 2026.

View on GitHub →

04 — Selected work

Featured projects.

🛰️

YOLOv8 Space Station Detector

★ Winner · IBA × Duality AI Challenge

Custom YOLOv8 model for safety-critical object detection in a simulated space station — Fire Extinguishers, Tool Boxes, Oxygen Tanks. Achieved mAP50 = 0.8675 with real-time inference through advanced data augmentation and hyperparameter tuning on synthetic data.

YOLOv8PyTorchOpenCVSynthetic Data

🩺

Multilingual Smart Clinic AI

GDG Kolachi · Build with AI Hackathon

Voice-enabled medical guidance agent supporting Sindhi, Pashto, Urdu, and English. ASR → LLM → TTS pipeline with safety checks, dialect handling, and emergency-protocol routing. Reduced system failures by 15% during real-time inference.

ASRLLMTTSMultilingual NLPPython

📊

Customer Propensity Model

★ Runner-Up · Evolve'25 Hackfest (NEDUET)

End-to-end ML pipeline on 11M+ vehicle insurance records achieving AUC 0.8793. Drove EDA, feature engineering, and class-imbalance handling. Co-presented results to industry and academic judges.

Scikit-learnXGBoostPandasFeature Engineering

🤖

Autonomous Hotel Workflow Agent

Agentic AI · Multi-step reasoning

Autonomous Python agents orchestrating end-to-end reservation flows. Multi-step reasoning with tool calling, state management, and workflow orchestration — eliminating manual staff dependencies for booking pipelines.

LangChainCrewAIAgentic AIPython

🏋️

Khifit — Gamified Fitness

Mobile · Health gamification

Gamified mobile fitness application promoting consistent health habits through challenge mechanics, streaks, and AI-personalized workout recommendations tailored to user activity patterns.

MobileGamificationAI Personalization

🍳

Recipe Generation Agent

Generative AI · Domain-specific

Focused generative AI system built on FLAN-T5-base for personalized meal planning and context-aware culinary instruction. Integrated structured recipe datasets for grounded outputs.

FLAN-T5Hugging FaceFine-tuning

View all projects on GitHub →

05 — What I do best

Core expertise.

Computer Vision

Fine-tuning vision transformers (DINOv2, ViT), training YOLOv8 object detection on custom datasets, and integrating CV into production pipelines.

Agentic AI & RAG

Building autonomous agents with LangChain, LangGraph, CrewAI. Designing RAG pipelines with ChromaDB, FAISS, and grounded LLM responses.

Multilingual NLP

Speech-to-speech systems supporting low-resource languages. ASR → LLM → TTS pipelines with safety filtering and dialect handling.

MLOps & Deployment

End-to-end model lifecycle: training, validation, CI/CD, containerization. Building production REST APIs around AI models with FastAPI and Flask.

06 — Technical arsenal

Skills & tools.

Languages

PythonSQLJavaScriptC++

AI / ML / Deep Learning

PyTorchTensorFlowScikit-learnKerasYOLOv8DINOv2Hugging FaceTransformersOpenCV

Agentic AI & LLM Orchestration

LangChainLangGraphCrewAIAutoGPTn8nZapierRAGPrompt Engineering

Core Techniques

CNNRNNANNFine-tuningModel TrainingValidation & TestingHyperparameter TuningFeature Engineering

Vector DBs & Retrieval

ChromaDBFAISSPineconeSentenceTransformers

Backend & APIs

FastAPIFlaskStreamlitGradioREST

Data & MLOps

PandasNumPyMatplotlibSeabornCI/CDDocker

Databases

PostgreSQLMySQLMongoDBSQLite

Tools & Platforms

GitGitHubLinuxVS CodeJupyterGoogle ColabKaggle

07 — Recognition

Achievements.

★

1st Place & ML Competition Winner · ZabLens AR campus navigation system

ZABFEST 2026 · SZABIST

🥇

Winner — Object Detection Challenge · Custom YOLOv8 model (mAP50 = 0.8675)

IBA × DUALITY AI

🥈

Runner-Up — AI/ML Dev Arena · Customer-propensity model on 11M+ records (AUC 0.8793)

EVOLVE'25 · NEDUET

🎓

Selected Trainee — Fully Funded Residency · 3-month diploma in Deep Learning & AI Communication

NAVTTC PMYSDP · IBA

09 — Beyond the code

Leadership.

Program Representative (PR)

BS AI Department · SZABIST

Elected liaison between faculty and 100+ AI students. Drive curriculum feedback and department-wide initiatives.

Student Council Member

SZABIST University

Spearhead technical awareness campaigns and co-organize university-wide hackathons.

Brand Ambassador

DataCrumbs

Successfully onboarded 300+ students into a specialized AI learning community through targeted outreach.