Pathik Ghugare

AI Researcher

Mumbai, Maharashtra, India5 yrs experience

AI EnabledAI ML Practitioner

Key Highlights

Expert in building AI-driven document processing solutions.
Proven track record in improving model accuracy significantly.
Strong experience in deploying scalable machine learning models.

Stackforce AI infers this person is a Machine Learning Engineer specializing in AI and Computer Vision solutions.

Contact

Skills

Core Skills

Machine LearningLarge Language Models (llm)MlopsComputer VisionNatural Language Processing

Other Skills

AWSAWS LambdaAmazon Web Services (AWS)Azure Machine LearningC (Programming Language)C++CI/CDCNNCVATData AnalysisData CleaningData CollectionData MiningData ScienceData Scraping

About

Hey there 👋 I’m Pathik, a machine learning engineer who’s all about building cool stuff with AI. From fine-tuning LLMs to making sense of complex document data, I’m always up for a challenge. Outside of code, I enjoy a good run and can never say no to a solid tech discussion. Let’s connect and talk ML, tech, or anything nerdy!

Experience

5 yrs

Total Experience

1 yr 3 mos

Average Tenure

1 yr 8 mos

Current Experience

Parspec

Machine Learning Engineer

Oct 2024 – Present · 1 yr 8 mos · Bengaluru, Karnataka, India · Hybrid

Built a multi-agent parsing pipeline with intelligent document classification and routing, achieving over 95% accuracy in structured data extraction.
Improved transformer models for text detection with token level manipulations, boosting accuracy from 89% to 95% across different domains.
Designed a classifications model combining embeddings and OCR responses, attaining 92% recall.
Engineered a variation detection system using GPT-4o with an LLM-as-Judge feedback loop to ensure annotation quality
Developed a lightweight table complexity classifier with latency as small as 300ms
Enhanced table parsing with CV-based preprocessing and code optimizations, improving performance by 50%.
Led a 2-member team to deliver end-to-end AI-based table extraction modules.
Partnered with platform teams to design CI/CD for ML models using GitHub Actions, ArgoCD, and OpenSearch for observability.
Deployed and scaled models across AWS ECS and GKE, ensuring fault-tolerant and reproducible workflows.
Benchmarked GPT-5, Claude, and Gemini for table parsing tasks, establishing LLM evaluation framework for future experiments

Data StructuresGPT-4Document ClassificationAWSCI/CDGitHub Actions+3

Pibit.ai

2 roles

Machine Learning Engineer

Jun 2023 – Oct 2024 · 1 yr 4 mos

Building DocumentAI at scale
Built the AI side of lossrun product from scratch
Built various Identity services using OpenAI GPT & Langchain, deployed it on AWS Lambda
Built MLOps pipelines for training and validating current models using Azure Machine Learning
Built AWS CodePipelines for model deployment on kubernetes cluster using helm charts
Working on handelling Incremental Data Drift using feedback loops
Running experiments using various object detection models, Multimodal architectures and LLMs
Built an OpenAIAssistant using GPT4, langchain and streamlit for data extraction from complex documents

OpenAILangchainAWS LambdaMLOpsAzure Machine LearningKubernetes+1

Deep Learning Intern

Aug 2022 – Jun 2023 · 10 mos

Improved existing object detector score from 0.62 to 0.74
Worked on OCR free document parsing and achieved an accuracy of 0.81
Built an auto annotation pipeline on CVAT
Integrated MLFlow with an object detector library
Built an internal dashboard for model metric visualisation and comparison using streamlit
Trained and deployed an object detection model for doing data extraction from identity documents

Object DetectionOCRCVATMLFlowStreamlitComputer Vision

Teaminup

Machine Learning Intern

Mar 2022 – Jul 2022 · 4 mos

Scraped internship and job data from the web using Selenium Python
Trained a word2vec model on skills to find out the similarity between two sets of skills and tags which was used in the recommendation system
Using the same word2vec model, wrote an API to generate similar skills and tags w.r.t. given skills and tags respectively
fine-tuned a Small BERT model in TensorFlow to predict roles from the description of a project with a validation accuracy of 78%.
Created a neural network in TensorFlow to predict skills from the set roles
Built a Flask API to deploy models as a web service

SeleniumWord2VecFlaskTensorFlowNatural Language Processing

Kj somaiya college of engineering, vidyavihar

Machine Learning research intern

Jun 2021 – Sep 2021 · 3 mos · Mumbai, Maharashtra, India

Collaborated with my teammates to do research on what can be done to recognize facial expressions where 10-20 research papers were read and a multi-task learning approach was finalized.
Built a Multi-task learning CNN architecture which was trained not just to recognize an expression but also other 3 facial properties such as age, gender, and race/ethnicity.
The proposed Multi-task learning approach gave an accuracy of 78% which was better than the traditional approach
Published our work on arXiv: https://arxiv.org/abs/2110.15028
Open-sourced our code for further work: https://github.com/sashankmvv/Emotion-recognition

Multi-task LearningCNNFacial Expression RecognitionMachine Learning