Available for opportunities
KoushikPulivarthi
AI Engineer  ·  LLM Systems  ·  AWS  ·  MLOps

Building production-grade AI systems — fine-tuned LLMs, RAG pipelines, and scalable model serving on AWS. I turn raw data into intelligent, deployed solutions that reduce latency, improve accuracy, and ship to production.

llm-pipeline · prod · us-east-1
0%
LLM Accuracy ↑
0%
Pipeline Efficiency
0%
Latency Reduced
0
AWS Services
Core Stack
🤖
LLM & GenAI
Language Models
Hugging FaceLangChainLlamaIndexRAG / FAISSFine-TuningPrompt Eng.BLEU/ROUGESentenceTransformers
☁️
Cloud · AWS
AWS Infrastructure
SageMakerBedrockS3Glue ETLEC2LambdaIAMIoT Core
ML & Training
Machine Learning
PyTorchTensorFlowXGBoostLightGBMScikit-LearnSHAPwandbMLflow
🔧
APIs & Serving
Model Deployment
FastAPIFlaskDockerKubernetesCI/CDQuantizationvLLMOllama
👁️
Computer Vision
Vision Systems
YOLOv8OpenCVMediaPipeReal-Time Detection
📊
Data Engineering
Data & Analytics
PandasNumPyDatabricksSnowflakeSQLSparkStreamlit
🛠️
MLOps & DevOps
Infrastructure
Git / GitLabJenkinsAnsibleTerraformGrafanaPrometheusn8nDVC
💻
Languages
Programming
PythonSQLJavaJavaScriptC++BashEmbedded C
Projects
01LLMAWS
End-to-End LLM Pipeline @ Toyota
Production LLM pipelines with advanced prompt engineering (few-shot, chain-of-thought, instruction tuning), RAG via FAISS vector databases, fine-tuned transformer models on industry-specific training datasets, and full evaluation framework with A/B testing and automated benchmarking.
Accuracy+30%
Efficiency+40%
Latency−35%
Reliability+25%
02RAGGenAI
RAG & Generative AI Systems
Production-ready RAG systems with semantic document search and generative Q&A. Dense vector retrieval via FAISS combined with transformer-based generation for accurate, grounded responses at scale.
RetrievalSemantic
InterfaceGradio
03LLM
Python Syntax Correction Tool
Interactive code correction powered by StarCoder2-7B and Llama-3.1-8B. Accepts buggy Python code and returns corrected output via real-time model inference using Hugging Face Transformers and Accelerate.
Models2 LLMs
SpeedReal-Time
04Computer Vision
Vehicle Detection & Speed Tracking
Real-time vehicle detection and speed estimation from video feeds using YOLOv8 and OpenCV. Computes speed via frame-to-frame displacement measurement and logs vehicle ID, speed, and timestamps for traffic analysis.
DetectionReal-Time
ModelYOLOv8
05ML · Finance
Stock Market Analysis Platform
AI-powered stock analysis platform with real-time financial data, RSI/SMA technical indicators, news sentiment analysis via OpenAI API, and YouTube video insights. Full interactive visualization dashboard.
APIs4 Sources
AnalysisReal-Time
06IoT · Safety
RF & Alcohol Detection Safety System
Real-time road safety system using RF and alcohol sensors to detect speeding and intoxicated driving simultaneously. Developed alert mechanisms for immediate driver and authority notification to enhance road safety.
SensorsMulti
ResponseReal-Time
🏗️ AWS IoT Data Pipeline
End-to-end sensor data pipeline on AWS Free Tier — IoT Core → S3 → Glue → SageMaker with XGBoost, SHAP explainability, and 24h forecasting.
IoT CoreSageMakerGlue ETLXGBoostSHAP
Skills & Expertise
AI / LLM Engineering
Prompt Engineering
95%
LLM Fine-Tuning
92%
RAG Systems
90%
Model Evaluation
88%
RLHF / Alignment
75%
AWS & Cloud
SageMaker
85%
S3 / Glue / EC2
82%
Lambda / IAM
78%
Bedrock
75%
Currently Learning / Using
LlamaIndexvLLMOllama LangGraphDSPyGroq Mistral APIAnthropic APIOpenAI API MLflowDVCWeights & Biases
ML & Data Engineering
Python / Pandas
95%
PyTorch
88%
XGBoost / LightGBM
85%
Databricks
72%
Snowflake
70%
DevOps & MLOps
DockerKubernetes FastAPICI/CD GitHub ActionsTerraform GrafanaPrometheus AWS SageMakerS3 / Glue BedrockLambda
Frameworks & Libraries
StreamlitGradioFlask KerasTensorFlowScikit-Learn SciPyYOLOv8OpenCV MediaPipen8nLangChain
Work Experience
Toyota
Aug 2025 – Present  ·  11 months total
AI Engineer
Current Role Nov 2025 – Present  ·  8 mos
+30%
Accuracy
−35%
Latency
+40%
Efficiency
Architected end-to-end LLM pipelines — improved response accuracy ~30% via few-shot chain-of-thought instruction tuning, RAG with FAISS vector databases, and Hugging Face fine-tuning on industry-specific training datasets.
Designed scalable data pipelines boosting preprocessing efficiency 40% — Python, pandas, NumPy; integrated modular workflows for dataset versioning batch processing and reproducible experimentation.
Engineered high-performance model serving infrastructure reducing inference latency 35% via FastAPI model quantization batching strategies and optimized request handling for real-time production workloads.
Built LLM evaluation & monitoring framework improving reliability 25%A/B testing BLEU/ROUGE semantic similarity metrics with automated benchmarking and statistical significance testing.
AI Engineer Intern
Internship Aug 2025 – Oct 2025  ·  3 mos
Developed LLM-powered features and fine-tuned models on domain-specific datasets to improve accuracy and reasoning capabilities.
Built workflows for data preprocessing, model training, evaluation, and continuous model improvement pipelines.
Conducted research on emerging LLM techniques and collaborated cross-functionally to implement AI-driven solutions across teams.
Education & Certifications
New York Institute of Technology
MS · Computer Science · New York, NY
Analysis of Algorithms · Database · Machine Learning · Deep Learning · Artificial Intelligence
R.M.K Engineering College
BE · Electronics & Communication · Chennai, India
Programming · Data Structures · Microcontrollers · Digital Electronics · VLSI Design
Python for Data Science — edX Programming for Everybody — Coursera AI For Everyone: Master the Basics — edX 🏆 3rd Place · Paper Presentation 🏆 Round 3 Finalist · IICC — AICTE & Coding Ninjas
Let's Build
Something.
Open to AI Engineer, ML Engineer, and LLM Engineer roles.
Especially interested in production AI systems, AWS-native ML, and LLM infrastructure.