Samir J.

Key Skills

Software

Other

Top Subject Matter

Spatial-temporal video anomaly detection (UCSD Ped1)

LLM-based IT helpdesk automation with RAG

Mental health risk classification from social media datasets

Top Data Types

Video

Image

Top Task Types

Prompt + Response Writing (SFT)

Classification

Freelancer Overview

AI/ML Intern — anomaly detection workflow using ConvLSTM autoencoder. Brings 3+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Internal, Proprietary Tooling, and Other. Education includes Master of Technology, University of Petroleum and Energy Studies (2026) and Bachelor of Technology, Bharati Vidyapeeth College of Engineering (2023). AI-training focus includes data types such as Video, Computer Code, and Programming and labeling workflows including Evaluation, Rating, and Prompt + Response Writing (SFT).

Labeling Experience

IT Support Agent — multi-agent RAG pipeline for incident resolution

OtherPrompt + Response Writing (SFT)

In the IT Support Agent project, Samir created a multi-agent RAG pipeline to generate contextual responses for IT incident resolution. He designed retrieval, troubleshooting, ticketing, and summarization components that work together to answer user queries in context. The pipeline also supports escalation decisions based on answer quality. • Implemented a semantic search and knowledge retrieval workflow over IT support documents. • Built Retrieval, Troubleshooting, Ticketing, and Summarization agents in a LangGraph setup. • Used Sentence Transformers and Hugging Face datasets for knowledge retrieval and response preparation. • Added scoring and intelligent escalation for query outcomes.

2026 - 2026

Machine Learning Project — mental health prediction and risk classification

OtherVideoClassification

In the mental health prediction project, Samir performed machine learning modeling on social-media derived data to classify mental health disorder risk. He implemented multiple algorithms and selected the best-performing model based on evaluation metrics. His focus was on building predictive capability from text-like social media inputs rather than manual annotation. • Conducted mental health risk analysis using social media datasets. • Trained and evaluated multiple ML models for predictive performance. • Selected Random Forest as the top model (reported 96% accuracy, 95% kappa). • Interpreted model outputs for disorder risk identification.

2025 - 2025

AI/ML Intern — anomaly detection workflow using ConvLSTM autoencoder

Video

During the IBM internship, Samir built and evaluated a ConvLSTM autoencoder for anomaly detection on video data. He used reconstruction error from the UCSD Ped1 workflow to flag suspicious events for visualization. His work centered on identifying and scoring anomalous segments for downstream interpretation of results. • Preprocessed UCSD Ped1 video frames for model input. • Detected anomalies via reconstruction error thresholds. • Visualized suspicious events in an intuitive output workflow. • Achieved strong anomaly detection performance (reported 97% AUC).

2025 - 2025

Education

Work History