Resume

πŸ‘‹ Hello, I’m

Rohan Patil

Building AI Systems
That Scale πŸš€

AI/ML Engineer with experience at Perplexity and Amazon, building production-grade LLM pipelines, RAG systems, and distributed ML infrastructure for real-world high-scale environments.

Resume

How I build systems

|

Selected Projects

Adaptive RAG Chatbot

Built an adaptive Retrieval-Augmented Generation (RAG) system that improves LLM reliability by dynamically selecting retrieval strategies based on query complexity. The system balances latency and accuracy by routing simple queries through lightweight retrieval while applying deeper contextual search and re-ranking for complex queries.

LangGraphQdrantFAISSMongoDBFastAPIGPT-4o

LENS β€” AI Image Intelligence

Developed a real-time multimodal AI system that processes images and generates contextual outputs across multiple modes including storytelling, humor, and analytical reasoning. The system leverages GPT-4o Vision with streaming responses to deliver an interactive and engaging user experience.

GPT-4o VisionNext.jsStreaming (SSE)Multimodal AI

Second Brain β€” Knowledge Graph

Built an AI-powered system that converts raw notes and text into structured knowledge graphs. The system extracts entities and relationships using LLMs and visualizes them as an interactive force-directed graph, enabling intuitive exploration of complex information.

GPT-4oD3.jsNext.jsKnowledge Graphs

Human Activity Recognition System

Built a lightweight human activity recognition system using 2D pose keypoints instead of raw video, enabling efficient sequence modeling with high accuracy. The system leverages temporal patterns using LSTM networks while significantly reducing computational overhead compared to RGB-based approaches.

TensorFlowLSTMOpenPoseFlask

Driver Drowsiness Detection System

Developed a real-time driver drowsiness detection system that monitors eye states using computer vision and deep learning. The system processes webcam input, detects facial regions, and classifies eye states to trigger alerts when fatigue is detected.

OpenCVCNNKerasPygame

Location Intelligence & Clustering System

Built a location intelligence system that analyzes geospatial and venue data to identify optimal neighborhoods. The system applies clustering techniques to group similar areas and provide insights for decision-making based on data patterns.

KMeansPandasScikit-learnGeospatial Analysis

Experience

Perplexity

AI/ML Engineer β€” Perplexity

June 2024 – Present Β· San Francisco, CA

  • β€’ Architected RAG pipelines integrating vector search + web indexing.
  • β€’ Built FAISS + Redis hybrid retrieval improving recall/precision tradeoff.
  • β€’ Optimized Triton GPU inference β†’ +25% throughput.
  • β€’ Designed LLM routing (on-device + cloud) for sub-second latency.
  • β€’ Improved factual consistency via ranking + citation pipelines.
  • β€’ Built evaluation systems tracking latency, accuracy, UX metrics.
  • β€’ Led 0β†’1 agentic AI features β†’ +18% engagement.
Amazon

AI/ML Engineer β€” Amazon

Oct 2019 – June 2023 Β· India

  • β€’ Built batch + streaming pipelines using AWS, Spark, Kafka.
  • β€’ Designed feature systems β†’ +30% faster data access.
  • β€’ Prevented training-serving skew in real-time ML systems.
  • β€’ Built Kafka + Spark streaming pipelines for low latency updates.
  • β€’ Orchestrated ML workflows with Airflow + SageMaker.
  • β€’ Built drift detection + monitoring datasets.
  • β€’ Reduced infra cost by ~15% via optimization.

Tools & Technologies

AI / ML

PyTorchTensorFlowScikit-learnNumPyPandasTime SeriesModel Evaluation

LLM / GenAI

RAGLangChainLangGraphOpenAI APIsEmbeddingsPrompt EngineeringSemantic SearchLLM Evaluation

Data Engineering

Apache SparkKafkaAirflowETL PipelinesParquetStreaming SystemsFeature Engineering

Infrastructure & Cloud

AWSKubernetesDockerSageMakerRedshiftGPU InferenceTriton ServerCI/CDPrometheusGrafana

Let’s Build Something Great πŸš€

I’m open to AI/ML Engineering roles, collaborations, and interesting problems. Feel free to reach out β€” I’d love to connect.

Β© 2026 Rohan Patil β€” Built with Next.js πŸš€