Shivam – Python, Typescript, React

This project focused on improving the quality, reliability, and reasoning capabilities of next-generation large language models used for software engineering tasks. The work involved evaluating AI-generated code, identifying weaknesses in reasoning and implementation, and providing structured human feedback to enhance model performance.

• Acted as a domain expert to evaluate state-of-the-art LLMs (including Claude’s and Gemini's unreleased models) for frontier coding capabilities;
• Stress-tested model-generated code for logic, efficiency, and complex edge cases;
• Provided critical RLHF (Reinforcement Learning from Human Feedback) for outputs in Rust, TypeScript, and Go;
• Authored comprehensive technical reports to document model hallucinations and architectural flaws;
• Refined the reasoning and overall accuracy of AI coding agents through structured benchmarking.

Developed and maintained a full-stack B2C SaaS platform focused on generating highly personalized astrological and geographic insights. The platform processes large-scale astronomical ephemeris datasets to calculate planetary positions with high precision in real time, then combines proprietary analytical algorithms with AI-powered text generation to transform complex astronomical data into customized user reports and assessments. The system was designed to handle significant data volumes, serving tens of thousands of generated reports while supporting the platform's rapid growth and commercial success.

• Architected a highly performant serverless application (Next.js, Vercel, PlanetScale) that successfully scaled to process over 65,000+ complex user reports;
• Engineered a precision data pipeline integrating the Swiss Ephemeris and NASA JPL DE431 data for real-time, topocentric coordinate calculations and Gaussian distribution mapping;
• Developed a dynamic text-generation engine to translate raw geographic vectors into deeply personalized "Psychological Diagnostics" and alignment protocols;
• Built interactive, data-heavy UI components to visualize geographic resonance, alignment scores (0-100%), and complex time-series transits;
• Owned the end-to-end commercial lifecycle, integrating Stripe for seamless paywalls, resulting in immediate profitability and securing Stripe Capital funding.

A real-time AI platform designed to improve collaboration and alignment in remote and hybrid teams by analyzing the emotional dynamics of meetings. The system processes audio and conversational data through a pipeline of large language models and custom machine learning models to generate post-meeting insights, including emotional trend analysis, team alignment metrics, and actionable summaries. The platform is currently used by a limited group of enterprise customers and focuses on helping organizations identify communication patterns and potential sources of team friction.

• Engineered a real-time, low-latency AI pipeline integrating multiple LLMs and custom machine learning models;
• Developed a proprietary prosody and facial expression model to detect conversational nuances (e.g., vocal frustration, high enthusiasm);
• Architected a system to generate automated "Post-Meeting Debriefs", calculating an aggregated "Alignment Score" based on relational data;
• Designed and built a highly reactive frontend dashboard (Next.js) featuring interactive time-series data visualizations (the "Emotional Arc");
• Translated complex, multi-modal sentiment data into intuitive and actionable insights for leadership and project managers;
• Owned the entire product lifecycle from initial architecture and training datasets to full-stack execution and sales conversion.

Contributed to the development of a high-scale digital healthcare and pharmacy platform that streamlines prescription management, appointment booking, and medication delivery. The platform connects patients with local pharmacies and healthcare providers through secure, real-time services. The project focused on building a scalable backend capable of handling sensitive healthcare data, integrating with external healthcare APIs, and leveraging Python-based AI microservices to optimize logistics, routing, and prescription data processing.

• Architected a robust backend ecosystem using Node.js and Express, designing high-throughput REST and GraphQL APIs to handle secure patient data and real-time appointment bookings;
• Designed a complex, highly efficient database architecture utilizing MongoDB geospatial queries and polygons to accurately map dynamic 30-minute delivery zones and match patients with local pharmacies;
• Developed custom Python AI microservices to handle predictive logistics routing and demand forecasting for the delivery dispatch engine;
• Implemented AI-driven Optical Character Recognition (OCR) and Natural Language Processing (NLP) pipelines in Python to automate the parsing of uploaded NHS prescriptions, drastically reducing manual data entry for pharmacy owners;
• Managed external tech partner integrations via robust webhook architectures, ensuring HIPAA/GDPR compliance across all backend services;
• Led a team of 8 developers, overseeing complex R&D tickets, sprint planning, and the scaling of cloud infrastructure to support 50k+ daily active users.