Palak P.

Key Skills

Software

Other

Top Subject Matter

Multimodal AI

Speech

Computer Vision

Top Data Types

Text

Document

3D Sensor

Image

Top Task Types

Classification

Prompt + Response Writing (SFT)

Question Answering

Function Calling

Freelancer Overview

Machine Learning Engineering Intern — stacked forecasting and deployment. Brings 2+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Other, OpenAI, and MLflow. Education includes Bachelor of Engineering, Nanyang Technological University (2025). AI-training focus includes data types such as Computer Code, Programming, and Medical and labeling workflows including Evaluation, Rating, and Classification.

Labeling Experience

MaintenancePilot — predictive-maintenance modeling and monitoring

3D SensorFunction Calling

Built a predictive-maintenance pipeline that uses sensor data to forecast air-leak failures 48 hours in advance. Productionized the pipeline with continuous monitoring for feature drift to maintain model/data reliability. Ensured robust deployment via CI/CD and model registry practices that support iterative training and updates. • Trained and deployed a LightGBM predictive-maintenance model on 1.5M rows of sensor data. • Set up FastAPI deployment in Docker on Hugging Face Spaces. • Used MLflow registry and CI/CD drift monitoring with Evidently and GitHub Actions. • Configured the pipeline to fail when feature drift exceeds a 50% threshold.

2026 - 2026

AeroOps Copilot — GraphRAG evaluation and QA

OtherDocumentQuestion Answering

Developed a live GraphRAG copilot over a turbofan-engine maintenance knowledge graph to support multi-hop question answering. Built parallel retrieval pipelines that compare graph-structured retrieval versus flat-chunk RAG to validate performance on hard technical maintenance questions. Designed an evaluation benchmark with hand-verified ground truth across multiple reasoning categories. • Shipped a full-stack GraphRAG copilot on Streamlit with Neo4j backend. • Built two retrieval pipelines (GraphRAG vs flat-chunk RAG) using LangChain and FAISS/BM25. • Created a 35-item benchmark with hand-verified ground truth across 7 reasoning categories. • Achieved 99% accuracy on multi-hop causal questions in the evaluation.

2026 - 2026

Associate Scientist Intern - Procter and Gamble

Image

Built and deployed an end-to-end computer-vision system to automate skin-tone measurement for P&G’s skin-imaging platform. Designed evaluation methods to select the best image-search approach for an AI perfume-design tool based on relevance, diversity, and palette. Created a CLIP-based labeling pipeline for a 10,000+ image scalp-health dataset and communicated results to international stakeholders. • Automated skin-tone measurement with strong detection accuracy and precision • Built an evaluation framework for comparing multiple image-search configurations • Auto-labeled a scalp-health image dataset by severity to reduce manual tagging • Presented trade-offs and results in stakeholder reviews to guide product decisions

2025 - 2025

AI Engineer — agentic outreach generation

TextPrompt + Response Writing (SFT)

Built and deployed an agentic system that generates personalized outreach by automating lead discovery and drafting targeted messages. Prototyped agentic and RAG workflows with pilot evaluations to determine which solutions were worth scaling. Conducted discovery conversations to translate business requirements into AI systems that could be scoped and shipped. • Implemented an agent to scrape leads and write personalized outreach. • Used LangChain/LangGraph with OpenAI GPT for generation workflows. • Ran pilot evaluations for agentic/RAG systems across business workflows. • Converted client discovery discussions into implementable AI solution scope.

2025 - 2025

Associate Scientist Intern — computer-vision pipeline and auto-labeling

OtherClassification

Automated computer-vision skin-tone measurement to generate reliable, reproducible measurements for a skin-imaging platform. Built evaluation and selection frameworks to compare image-search configurations and choose the best approach based on objective scoring. Created an image-based labeling pipeline to assign severity labels to a large dataset, reducing manual tagging effort. • Automated skin-tone measurement via an end-to-end computer-vision system. • Designed an evaluation framework scoring relevance, diversity, and palette across four configurations on a 50,000-image library. • Built a CLIP-based auto-labeling pipeline for 10,000+ scalp-health images by severity. • Communicated results and trade-offs in stakeholder reviews and incorporated feedback into iterations.

2025 - 2025

Show more

Education

Work History