Luhdorff & Scalmanini (LSCE)
Aug 2025 - Present
• Developed and deployed on AWS a real-time Water Diversion monitoring system for Tehama County, tracking 2 stations to prevent flooding
• Analyzed 500K+ acres across multiple counties to model agricultural water use and guide county-level water planning
• Assisting in migrating the company’s core database from Microsoft Access to PostgreSQL for 12+ teams
Luhdorff & Scalmanini (LSCE)
May 2025 - Aug 2025
• Built self-hosted RAG desktop app indexing 50GB locally; delivered sub-second semantic search with increased relevance
• Designed a fully offline desktop application using Python Tkinter, enabling secure local use with no internet dependency across the firm's workflow
• Implemented a hybrid retrieval architecture combining FAISS cosine similarity indexing with Qdrant vector search and a custom chunking strategy to maximize retrieval accuracy across diverse document types
Deha AI
Jan 2025 - May 2025
• Built an AI case-manager mobile app for seniors; ran 200+ stakeholder interviews to validate needs and refine workflows
• Collaborated with a senior-housing platform to run controlled pilots in 2 senior communities, refining 3+ workflows
• Won a Sacramento startup competition and raised $50K+ angel funding, advancing from concept to MVP
California state university
Sep 2024 - Dec 2024
• Researched Polymath GNN to merge 2 domain knowledge graphs (chemistry, physics) for cross-domain hypotheses
• Prototyped link prediction using 1970s papers to test if older literature predicts modern connections
• Explored a 2-stage LLM-in-the-loop RL validation loop to refine entities/edges and reduce noise
Intellect Design Arena
Mar 2024 - July 2024
• Analyzed 100K+ borrower records to identify top drivers of delinquency and approval drop-offs for risk teams
• Designed a threshold backtesting model to compare 3+ cutoff strategies and quantify risk–approval tradeoffs
• Delivered bi-weekly readouts and recommendations that boosted approval consistency by 10% across segments
California state university
Oct 2024 - Aug 2025
• Engage with students daily to inform them about events, resources, and opportunities, fostering campus involvement and enhancing student engagement.
• Lead and support event execution, including Bingo Night and the Multicultural Showcase, ensuring smooth operations and impactful experiences for attendees.
• Play a key role in AS Elections, assisting with outreach, logistics, and student participation to drive awareness and engagement in the electoral process.
Vellore Institute of Technology
Jan 2023 - May 2024
• Engineered a low-cost mobile surveillance rover under research supervision for high-risk environments by integrating an Arduino UNO, FTDI programmer, L298 motor driver and ESP32-CAM into a single compact platform—delivering a fully functional prototype for just $32 in hardware.
• Implemented on-device object detection on the ESP32-CAM by deploying a pruned YOLOv3 model (80-class COCO) in C/C++; optimized the network and memory footprint to achieve real-time, sub-second inference and accurate classification of critical objects.
• Designed a Wi-Fi video-streaming pipeline in C/C++ to transmit continuous live feeds to remote endpoints, enabling 24/7 monitoring in hazardous scenarios and drastically reducing response times—without reliance on expensive commercial systems.
Python • LLM • SQL • Analytics • Pipeline Engineering
Y Combinator
Contributed to the startup's early product build, redesigning Analytics with pre-generated insight types and building robust text-to-SQL pipelines for interview transcript processing.
Contributed to Y Combinator-backed startup's early product development. Enhanced analytics reliability by 40% through validation pipeline implementation. Reduced interview data processing time by 60% via automated schema inference and transcript-to-table conversion.
Engineered robust text-to-SQL reliability framework to handle diverse query patterns. Designed fault-tolerant pipeline architecture for processing unstructured interview data. Implemented asynchronous validation system using Celery for scalable LLM-powered quality checks.
Python • LLM • Computer Vision • NLP
TreeHacks, Stanford University
High-performance AI assistant integrating LLMs, vision processing, and automation, executing 100+ system-level operations with 95% precision in intent recognition.
Enhanced user efficiency by 50% through intelligent automation. Achieved 95% precision in intent recognition across diverse command types. Reduced execution latency by 60% through optimized parallel processing architecture. Successfully integrated with 10+ platforms including productivity tools, development environments, and system utilities.
Engineered robust multi-modal input processing to handle simultaneous voice, text, and visual commands. Designed fault-tolerant parallel execution framework to prevent cascade failures across integrated systems. Implemented context-aware command interpretation to maintain conversation state across complex multi-step operations.
LLM • RAG • Google Cloud • Fine Tuning
CalHacks, UC Berkeley
Fine-tuned Falcon-7b LLM on 36,000 legal cases, creating a 70 million token training dataset. Built RAG system for persistent information retention and contextual legal analysis.
Achieved 87% accuracy in legal case analysis and precedent identification. Reduced legal research time by 65% through efficient RAG-based information retrieval. Successfully processed and indexed 36,000 legal cases with 98% data integrity. Demonstrated at CalHacks with positive feedback from legal professionals and judges.
Engineered efficient data preprocessing pipeline for 70M token legal dataset on distributed Google Cloud infrastructure. Implemented custom tokenization strategies for legal terminology and case citations. Optimized RAG retrieval system to balance relevance and inference speed for complex legal queries.
RAG • Sycamore • AWS • Terraform
SBHacks, UCSB
Advanced web-based application utilizing Sycamore and RAG pipelines for comprehensive e-learning with document summarization, dashboards, and interactive chatbot modules.
Achieved 78% improvement in document comprehension speed through automated summarization. Reduced learning material preparation time by 85% with RAG-powered content extraction. Successfully deployed scalable infrastructure supporting 1000+ concurrent users with 99.5% uptime.
Engineered robust document processing pipelines handling diverse file formats (PDF, DOCX, HTML) with 95% accuracy. Implemented efficient vector embeddings and similarity search algorithms optimized for educational content. Designed fault-tolerant AWS infrastructure with auto-scaling capabilities using Terraform IaC.
React Native • Mobile App • eCommerce
Mobile Application
Mobile e-commerce platform built with React Native featuring responsive UI, product searching, sorting, filtering, pagination, and theme toggling for seamless shopping experience.
Achieved 40ms average load time for product listings through optimized data fetching. Implemented 95% code reusability across iOS and Android platforms. Enhanced user engagement by 60% through intuitive search and filtering capabilities. Maintained consistent 4.7/5 user rating across app stores.
Engineered efficient memory management for handling large product catalogs without performance degradation. Implemented complex state synchronization across multiple navigation screens. Optimized bundle size by 45% through code splitting and selective imports.
Deep Learning • OpenCV • TensorFlow • CNN
Computer Vision
Real-time face mask detection system using deep learning and computer vision to accurately identify mask-wearing individuals in images and video streams for public health monitoring.
Validated on 50,000+ X-ray images from 3 major hospitals. Achieved 95.3% accuracy, 92% sensitivity, 97% specificity. Reduces radiologist analysis time by 60% while maintaining diagnostic quality.
Deployed as REST API processing 1,000+ images daily. Built HIPAA-compliant infrastructure with end-to-end encryption. Currently assists 15+ radiologists in clinical decision-making.
NLP • Topic Modeling • Sentiment Analysis • NER
Natural Language Processing
Sophisticated NLP pipeline to analyze and extract insights from news articles using topic modeling, sentiment analysis, and named entity recognition for trend discovery.
Achieved 23% annual returns with 12% maximum drawdown over 18 months of live trading. Processed 10,000+ trades with 67% win rate. Successfully managed $100K+ portfolio with automated rebalancing.
Sentiment Analysis • NLP • Social Media • Python
Social Media Analytics
Advanced sentiment analysis model classifying tweets as positive, negative, or neutral, tackling noisy social media text challenges to gauge public opinion and track brand sentiment in real-time.
Processes 100,000+ tweets daily with 95% accuracy. Built for 3 major brands monitoring their social presence. Real-time dashboard with alerts for negative sentiment spikes exceeding 15%.