Internships

Machine Learning Intern — Acmegrade Apr 2025 – Jun 2025 · Remote · Ghaziabad, India

• Built a Cancer Prediction model achieving ~90% accuracy for early-stage detection using supervised classification and feature engineering pipelines.
• Developed a Music Recommendation system using collaborative filtering, achieving 85% accuracy in playlist personalization.
• Optimized data preprocessing workflows, reducing model training time by 25% through parallelized transformations and efficient data loading.

Skills: Python · Scikit-learn · Collaborative Filtering · Statistical Data Analysis · Feature Engineering

Machine Learning Intern — Codec Technologies India Jul 2024 – Sep 2024 · Remote · Ghaziabad, Uttar Pradesh, India

• Built an NLP-powered Resume Analysis tool with ATS compatibility scoring, improving candidate shortlisting accuracy by 40%.
• Designed a skill extraction pipeline using text processing and domain classification to compute job-fit scores against industry benchmarks.
• Performed exploratory data analysis on resume datasets to identify patterns in skill gaps and domain fit distributions.

Skills: Python · NLP · EDA · Streamlit · ATS Scoring

RLHF Evaluator & Annotator — Mercor | Project Neon Origin Phase 2B   2026– Present · Remote

• Applied ICON-REAL framework to evaluate LLM outputs across safety-critical and adversarial scenarios
• Scored model responses on 1–5 quality rubrics covering instruction following, coherence, and refusal accuracy
• Identified jailbreak attempts including fictional framing and roleplay-based manipulation tactics
• Performed structured annotation workflows using SuperAnnotate on Project Neon Origin Phase 2B

Skills: RLHF · LLM Evaluation · AI Safety · Prompt Analysis · Data Annotation · SuperAnnotate

AI/ML Projects

ML Experiment AutoPilot (In Progress) Multi-Agent AI System for Automated ML Experimentation

• Architecting a multi-agent orchestration system using the Anthropic Claude API, where specialized sub-agents handle dataset analysis, model selection, hyperparameter tuning, and result evaluation autonomously.
• Implementing agent memory, prompt contracts, and a tiered token cost strategy (Haiku/Sonnet/Opus) to optimize inference spend across pipeline stages.
• Integrating MLflow for experiment tracking and Streamlit for a real-time experiment dashboard with run comparison and metric visualization.

Skills: Claude API · Multi-Agent Systems · MLflow · Streamlit · Python

CrowdMind AI (In Progress) Real-Time Crowd Density Estimation & Risk Assessment System

• Built a real-time people counting prototype using YOLOv8n, HOG+NMS, and motion-based detection with OpenCV, addressing undercounting challenges in dense crowd scenarios.
• Integrating CSRNet with JHU-CROWD++ dataset for high-density crowd estimation using dilated convolutions.
• Designing a multimodal CNN+MLP fusion architecture combining learned visual density features with engineered spatial statistics for improved prediction robustness.
• Developing risk assessment overlay with density heatmap visualization for real-time safety monitoring.

Skills: YOLOv8 · CSRNet · OpenCV · PyTorch · Computer Vision · Deep Learning

AI Study Buddy (Completed) Local-First Productivity Monitoring & Conversational Analysis System

• Built an automatic laptop activity monitor that captures active application, window title, and idle time at regular intervals — eliminating manual time-tracking entirely.
• Designed a session aggregation engine that merges raw activity events into classified sessions (Focus / Distraction / Neutral / Away) and computes behavioral metrics including focus streaks, con frequency, and app usage distribution.
• Integrated the Grok API as a local LLM insight layer — structured behavioral summaries are passed to the model for natural language productivity analysis, keeping raw activity data on-device.
• Delivered insights through a Telegram chatbot interface supporting /today and /week commands alongside free-form productivity queries.

Skills: Python · Grok API · SQLite · Telegram Bot API · Behavioral Analytics

Agentic Honeypot (Completed) Autonomous AI-Driven Threat Interaction & Behavior Analysis System

• Designed decoy service endpoints that simulate vulnerable infrastructure to attract and log real attacker interactions.
• Built a behavior logging pipeline that structures attacker interaction data for downstream analysis — capturing session patterns, payload types, and interaction sequences.
• Integrated an LLM-based reasoning layer that classifies attacker intent, generates adaptive responses, and scores threat risk in real time.
• Focused on studying attacker behavior patterns rather than simply blocking — enabling threat intelligence collection from live interaction data.

Skills: Python · LLM Reasoning · Cybersecurity · Agent Architecture · Threat Analysis

AniMind (Completed) Con Anime Persona Conversational AI

• Built a character-consistent conversational AI system with persona conditioning prompts and conversation memory tracking across sessions.
• Implemented context window management and structured response generation pipeline to maintain behavioral consistency in long conversations.
• Explored token optimization strategies and conversation state architecture as a practical introduction to GenAI system design.

Skills: Python · GenAI · Prompt Engineering · Context Management · LLM APIs

CrimeScope AI (Completed) ML-Based Crime Prediction & Geospatial Visualization System

• Designed a crime prediction system that forecasts crime types by location using classification models trained on historical incident data.
• Built interactive heatmap visualizations and dashboards in Streamlit to support pattern analysis for law enforcement and researchers.

Skills: Python · Streamlit · Scikit-learn · Geospatial Analysis

Krishi AI (Completed) AI-Driven Agricultural Assistant for Crop Disease Prediction

• Designed a deep learning-based crop disease detection system that classifies plant conditions from images and provides actionable farming recommendations.
• Structured the project as a research paper submission, covering dataset curation, model architecture decisions, evaluation metrics, and deployment considerations.

Skills: Deep Learning · Computer Vision · Python · Research Paper