Harshit Singh

Founding Team US LLMs || Fine Tuning || Cloud Infrastructure

Developed a custom Python interpreter and AI compiler that automated complex data workflows, eliminating 95% of manual spreadsheet tasks and scaling to 100K+ records with real time accuracy Engineered agentic AI pipelines enabling LLMs to autonomously decide which APIs and services to invoke in real time, optimizing inference speed and system reliability Implemented LSTM-based demand forecasting that aligned production with sales patterns, cutting daily waste from 12% to 4% while maximizing operational efficiency

Led development of a Retrieval-Augmented Generation pipeline, integrating Elasticsearch, Neural Seek, and a fine-tuned Granite 13B model to boost chatbot accuracy from 0 to 96% and cut manual data handling by 80%. Optimized data infrastructure and MLOps workflows, using B-Tree indexing, partitioning, and automated Python/RMAN scripts for snapshots and backups—improving platform performance by 30% and slashing downtime by 60% for a service reaching 1.6 million+ residents.

Built and benchmarked trading signal models and ETL pipeline, comparing Random Forest and XGBoost to boost classification accuracy by 25%, and architecting a Python/SQL/AWS (EC2, S3, SageMaker) workflow to process over 1 million data points daily. Enhanced LSTM time-series forecasting, integrating bidirectional layers and attention mechanisms, then cutting inference latency by 15% through batch processing and selective pruning.

Revamped drug recommendation engine, applying Bayesian-optimized CART and JAX-powered parallel inference to lift prediction accuracy by 15% and accelerate classification throughput by 7% on 1M+ patient records. Engineered a Python/NetworkX graph pipeline to automate symptom-to-diagnosis-to-drug mapping, cutting manual analysis time by 20%.

Architected and developed the Aarush SRM festival website using the MERN stack, designing RESTful APIs with Node.js/Express, building React-based UIs, and deploying Docker-containerized services on AWS EC2—supporting 10,000+ daily visitors with zero downtime. Built a Progressive Web App for participant onboarding, leveraging React, Firebase Authentication, Firestore, and Cloud Functions to automate registration and event scheduling workflows, slashing manual processing by 70% and enabling real-time status updates.

Built and scaled dynamic websites for Inception Wave using the MEAN stack, partnering with designers and product owners to deliver responsive, feature-rich applications on an agile cadence. Mentored and led hands-on MEAN workshops for new team members, fostering cross-functional collaboration, accelerating onboarding, and equipping freshers to contribute production-ready code within weeks.

CS 411 Database Systems CS 412 Data Mining CS 440 Artificial Intelligence CS 435 Cloud Networking CS 467 Social Visualization CS 598 ML and Data Systems CS 568 User-Centered Machine Learning CS 598 Deep Generative Models