AI/ML Engineer
Akshat Tayal

Building Systems that Learn, Adapt & Scale

I build and deploy intelligent systems — from data to production.

About Me

Building real-world AI systems, not just models

I’m a Computer Science student focused on building real-world AI systems — from experimentation to deployment. My work spans intelligent document systems, computer vision, and scalable ML pipelines. I enjoy taking ideas beyond notebooks and turning them into usable, production-ready solutions. I care about performance, scalability, and practical impact — not just model accuracy.

I focus on solving problems end-to-end — from data processing and model design to APIs and deployment. Whether it’s optimizing inference, building retrieval systems, or integrating AI into real applications, I aim to create systems that actually work in the real world.

Modelsthat Work
Codethat Scales
Systemsthat Last
local_llm.py
# local_llm.py
from transformers import pipeline

# Load local LLM
pipe = pipeline(
    "text-generation",
    model="mistralai/Mistral-7B-Instruct-v0.1",
    device_map="auto"
)

# Run inference
response = pipe("Explain AI in simple terms",
 	 	  max_new_tokens=50,
 	 	  temperature=1.2)

print(response[0]["generated_text"]) 

Featured Projects

AI systems I've built from scratch — moving beyond tutorials to solve real-world problems.

IRS – Intelligent Retrieval System

Featured

Built a Retrieval-Augmented Generation (RAG) system to query PDFs using semantic search and LLMs, enabling context-aware and source-grounded answers.

Semantic SearchFAISS Vector DBConversational Memory
LangChainFAISSHuggingFaceLLaMA 3Streamlit

Self-Pruning Neural Network

Featured

Designed a neural network with learnable gates to dynamically remove redundant connections during training, achieving high sparsity without accuracy loss.

85% SparsityNo Accuracy DropLearnable Gates
PyTorchCIFAR-10Sparse Training

SRE Monitoring Pipeline (IaC)

Featured

Built an Infrastructure-as-Code system to deploy and configure a monitoring stack using Terraform and Ansible with Prometheus, Grafana, and Alertmanager.

Terraform AutomationCI ValidationDocker-Based Stack
TerraformAnsiblePrometheusGrafanaDocker

Fault Management (FM) NBI Flow

Featured

Built a production-grade REST API simulating a Network Management System (NMS) Northbound Interface for real-time fault detection, logging, clearing, and auto-simulation of network events using Spring Boot and H2 database.

Auto Fault SimulationREST API DesignSwagger Docs
JavaSpring BootH2 DatabaseJUnit 5Swagger/OpenAPIMaven

Real-Time Image Sharpening System

Developed a lightweight image sharpening model using knowledge distillation, achieving real-time performance with significant reduction in model size.

10x Faster Inference90% Parameter ReductionReal-Time Processing
PyTorchVGG19Knowledge DistillationOpenCV

Technical Skills

Technologies and systems I use to build production-ready AI applications

🔗

LLM / GenAI

RAG PipelinesLangChainFAISSHuggingFaceLLM Deployment
🤖

AI / ML

PyTorchScikit-learnOpenCVYOLOModel OptimizationKnowledge Distillation
🐍

Languages

PythonC++JavaScriptTypeScriptSQL
🌐

Web

ReactNext.jsFastAPIREST APIsTailwind CSS
🛠

Tools & DevOps

GitDockerLinuxGitHub ActionsJupyter
☁️

Cloud

AWS (S3, EC2)StreamlitHuggingFace SpacesGoogle Colab

Certifications

Continuous learning in a fast-paced field.

AI/ML

AWS Certified AI Practitioner

Amazon Web Services2026
Cloud

AWS Certified Cloud Practitioner

Amazon Web Services2026
AI/ML

Oracle Generative AI Professional

Oracle2025
AI/ML

Oracle AI Foundations Associate

Oracle2025
Data Science

Introduction to Data Science

Cisco Networking Academy2026

Achievements

Milestones and recognitions along the journey.

Award

Hackathon Winner

Award2025
Award

IIT Delhi Hackathon — Runner Up

Award2024
Scholarship

Performance-Based Scholarship Recipient

Scholarship2024

Experience

Where I've made an impact.

Software Engineering Intern

Intel Unnati Industrial Training Program
May 2025 – July 2025
  • Compressed VGG19 to 50% size using knowledge distillation while preserving 95%+ PSNR/SSIM
  • Built multi-threaded real-time inference pipeline with GPU FP16 acceleration (sub-second latency)
  • Added structured logging + telemetry for production observability

Let's build something meaningful

From ideas to production-ready systems.

Get in Touch

Have a project in mind or just want to chat about AI? I'd love to hear from you.