█████╗ ████████╗██╗  ██╗ █████╗ ██████╗ ██╗   ██╗ █████╗ 
██╔══██╗╚══██╔══╝██║  ██║██╔══██╗██╔══██╗██║   ██║██╔══██╗
███████║   ██║   ███████║███████║██████╔╝██║   ██║███████║
██╔══██║   ██║   ██╔══██║██╔══██║██╔══██╗╚██╗ ██╔╝██╔══██║
██║  ██║   ██║   ██║  ██║██║  ██║██║  ██║ ╚████╔╝ ██║  ██║
╚═╝  ╚═╝   ╚═╝   ╚═╝  ╚═╝╚═╝  ╚═╝╚═╝  ╚═╝  ╚═══╝  ╚═╝  ╚═╝

██╗    ██╗ █████╗ ██╗      █████╗ ██╗    ██╗ █████╗ ██╗     ██╗  ██╗ █████╗ ██████╗ 
██║    ██║██╔══██╗██║     ██╔══██╗██║    ██║██╔══██╗██║     ██║ ██╔╝██╔══██╗██╔══██╗
██║ █╗ ██║███████║██║     ███████║██║ █╗ ██║███████║██║     █████╔╝ ███████║██████╔╝
██║███╗██║██╔══██║██║     ██╔══██║██║███╗██║██╔══██║██║     ██╔═██╗ ██╔══██║██╔══██╗
╚███╔███╔╝██║  ██║███████╗██║  ██║╚███╔███╔╝██║  ██║███████╗██║  ██╗██║  ██║██║  ██║
 ╚══╝╚══╝ ╚═╝  ╚═╝╚══════╝╚═╝  ╚═╝ ╚══╝╚══╝ ╚═╝  ╚═╝╚══════╝╚═╝  ╚═╝╚═╝  ╚═╝╚═╝  ╚═╝

ai. robotics. everything in between...

looking for full time roles in robotics and ai research

email github linkedin huggingface blogs

Experience

Chewy Inc., Bellevue, WA — AI Engineering Intern (Jun 2025 – Aug 2025)
  • Architected an LLM-powered intent inference system analyzing 5M+ daily events, mapping user shopping journeys into structured behavioral states; recognized as Runner-up & "Most Innovative Use of AI" by leadership.
  • Designed a memory store with billions of entries persisting user intents and behavioral context, enabling downstream agents to deliver personalized experiences.
  • Built real-time behavioral pattern detection to classify journey stages with confidence scoring, replacing aggregate analytics with user-level insights.
  • Developed an autonomous campaign analysis agent leveraging stored intents for root-cause analysis; reduced manual analysis time by ~80% via Slack/Jira integration.
  • Cut inference costs ~60% by batching requests and mixing optimized open-source and vendor models.

Projects

Trip.AI [2024]
LLM-powered travel itinerary generator using Groq API. Strict JSON schema validation, dynamic itinerary modifications, cross-checking against Google and Wikipedia data, Leaflet maps integration, and per-person budget breakdown.
github
Cover Letter Generation with LLaMA [2024]
Fine-tuned LLaMA 3.2-3B for specialized cover letter generation using LoRA. Implemented efficient 4-bit quantized training pipeline using unsloth. Available on Hugging Face.
huggingface
EasyQuizzes [Cal Hacks 11.0, 2025]
AI-powered study assistant built in under 36 hours. Integrated OCR with LLaMA 3.2-11B-vision for document understanding. Built a RAG system with ChromaDB vector embeddings for question generation.
github
Deep Learning Flood Alert System [Patent Pending, 2024]
TensorFlow-based flood prediction with 120-hour advance forecasting. Achieved RMSE 0.7253 mm, MSE 0.5261 mm, MAE 0.1709 mm. Used Haversine formula to select the closest regional model based on user location. Deployed on Raspberry Pi with a Flutter mobile app for real-time alerts.
Isaac Gym Robotic Manipulation [2024]
Trained LeRobot SO-100 robotic arm in NVIDIA Isaac Gym using imitation learning from its paired arm for autonomous pick-and-place tasks. Combined learned policies with ROS-based fallback heuristics to ensure task completion under uncertainty. Integrated Gemini Vision API for real-time object detection and built a Flask + React dashboard for live monitoring.
github
Linux Systems Programming [2023]
Developed a kernel-level RLE compression driver with custom IOCTL interface. Implemented comprehensive error handling and memory management. Optimized read/write operations and designed system call hooks for seamless interaction.
github

Ongoing Projects

Gator Tutor Ongoing – SFSU Software Engineering Capstone
  • Built and maintained the FastAPI backend, structuring endpoints with Pydantic models for validation, serialization, and schema enforcement.
  • Designed and optimized the MySQL database with SQLAlchemy ORM, handling relational mapping for users, tutors, sessions, messages, and analytics.
  • Deployed and managed the app on Linux (Ubuntu 22.04) AWS EC2 servers, including environment setup, process control, and file system organization.
  • Automated manual deployments via SCP and tar archives, ensuring reproducible builds and version tracking without external CI/CD tools.
  • Configured NGINX for reverse proxy, static content delivery, and secure API routing between React and FastAPI services.
  • Implemented a search/filter system combining categorical filters with SQL %LIKE queries for accurate tutor discovery.
  • Managed database migrations, seeding, and backups, ensuring schema consistency across development and production.
  • Oversaw integration, debugging, and performance profiling to maintain backend reliability and scalability across milestones.
Adaptive Reinforcement Learning for Dynamic ROI Selection in rPPG (Thesis) 2025–2026
  • Building a PPO-based RL policy to dynamically weight facial regions for remote PPG signal extraction.
  • Using unsupervised signal quality metrics (SNR, spectral peak stability, inter-ROI coherence) as reward signals, replacing static geometric heuristics.
  • Training on public datasets (MMPD, V4V, PURE); benchmarking against R2I-rPPG and SAMC baselines.
  • Implementing a Kalman-filtered reward baseline for stability under motion and illumination variation.

Research

Deep Learning Flood Alert System [Patent Pending, 2024]
IoT and deep learning system combining LSTM networks with edge computing over Raspberry Pi for real-world environmental monitoring.
Credit Card Fraud Detection [Publication, 2023]
"A Comparative Analysis of Machine Learning Techniques for Credit Card Fraud Detection." Evaluated ensemble methods (XGBoost, Random Forest), traditional ML (SVM, Logistic Regression), and anomaly detection techniques. Implemented a comparison framework using scikit-learn and Python.

Hackathons & Competitions

Llama Hackathon — Cerebral Valley × Meta [2025]
Team project: WikiFace, an open-source platform inspired by DeepWiki and tailored for Hugging Face. Uses Llama's extended context window to present fine-grained information about models (papers, architecture trees, tokenization and implementation notes) with a Llama-powered chatbot interface.
Cal Hacks 11.0 [36 Hours, 2025]
Built EasyQuizzes: ReactJS frontend with ShadCN UI and FastAPI backend. Leveraged Groq, ChromaDB, and Gemini APIs for LLM-based question retrieval and summarization.
COEP Neurohacks [48 Hours, 2023]
English-to-Hinglish translator using a PyTorch-based NLP pipeline. Lightweight Gradio UI and efficient CPU inference optimized for hackathon constraints.

Robotics & Edge

• Isaac Gym → ROS: imitation-learning policies for SO-100 with fallback heuristics for reliability
• Edge deployment and quantization: TensorFlow Lite on Raspberry Pi; on-device alerts and monitoring

Technical Stack

ML/AI: PyTorch, TensorFlow, Hugging Face, ChromaDB, LLMs, RAG, TTS/STT Models

Programming: Python (NumPy, Pandas, Matplotlib, FastAPI), C++, MATLAB, SQL, JavaScript (React, ShadCN)

Tools & Platforms: Linux, Git, AWS (Basic), Google Coral, Raspberry Pi

Education

M.S., Data Science & AI | San Francisco State University | Expected Jul 2026

GPA: 4.0/4.0
Relevant coursework: Operating Systems, Natural Language Processing, Probability & Statistics, Pattern Analysis, Analysis of Algorithms

B.Tech., Electronics & Communication | Pune University | 2020-2024

GPA: 8.2/10.0
Relevant coursework: Data Structures & Algorithms, Embedded Systems, Signal Processing, IoT Development

Contact

Looking for full-time opportunities starting May 2026 in the ML/AI domain. Feel free to reach out if you want to discuss AI, collaborate on projects, or talk about opportunities.

atharvawal27@gmail.com | San Francisco, CA