Hi, I'm Samarpartap Passey

Sage

Adaptive AI Interview Prep Agent

An agentic AI oral examiner that conducts adaptive, voice-based mock interviews tailored to specific roles, experience levels, and interview types — then delivers structured readiness reports.

The production interview runs entirely through Vapi — GPT-4o handles question generation, answer evaluation, and adaptive follow-up logic end-to-end via a dynamic system prompt built at session initialization. A LangGraph agent architecture was built as the original structured implementation and remains in the codebase as a reference — defining a five-node state machine with explicit routing logic and deterministic decision-making. GPT-4o-mini handles structured tasks off the live path: breaking the role into three topic areas on initialization, and generating the final evaluation report with per-area readiness scores, weak moment analysis with exact quotes, and specific improvement suggestions.

Technologies Used:

Key Features:

• ✓LangGraph Agent Architecture: Five-node stateful graph with deterministic routing — no LLM call in the decision node
• ✓Vapi Voice Layer: WebRTC voice pipeline with VAD, echo cancellation, and real-time transcription via Vapi SDK
• ✓Dual-Model Design: GPT-4o for conversational interview, GPT-4o-mini for structured evaluation tasks
• ✓Adaptive Decision Loop: Drills deeper on weak answers, follows up on medium ones, advances on strong responses
• ✓Role-Tailored Interviews: Dynamically generates 3 topic areas per role, level, and interview type combination
• ✓Structured Evaluation Reports: Readiness scores, per-area breakdowns, and weak moment analysis with exact quotes
• ✓iOS Safari Workaround: ReadyScreen interstitial resolves getUserMedia gesture-chain requirement on mobile
• ✓Production Deployed: CI/CD pipeline on Vercel with Clerk auth gate after first free session

Skills Highlighted:

Finflow

AI-Powered Financial Analytics Platform

A production-ready full-stack web application that revolutionizes personal finance management through AI-powered transaction categorization, intelligent analytics, and mobile-responsive design.

Finflow processes financial transactions using OpenAI's GPT-4 API to automatically categorize expenses, identify spending patterns, and detect anomalies. The platform features a secure multi-user system with JWT authentication, adaptive CSV/Excel parsing for multiple Canadian banks (RBC, TD, Amex, Scotiabank), and an interactive dashboard with real-time data visualizations. Built with modern architecture principles, the codebase was refactored from 800+ to 350 lines while maintaining full functionality, demonstrating clean code practices and modular component design.

Technologies Used:

Key Features:

• ✓AI-Powered Categorization: Utilizes OpenAI GPT-4 for intelligent transaction classification
• ✓Adaptive Bank Parsing: Supports multiple Canadian bank statement formats with automatic data extraction
• ✓Anomaly Detection System: Identifies suspicious transactions, duplicate charges, and spending outliers using statistical thresholds
• ✓Recurring Charge Detection: 95%+ accuracy in identifying subscriptions and regular payments
• ✓Interactive Dashboard: Real-time data visualizations with spending breakdowns and budget tracking using Recharts
• ✓Mobile-Responsive Design: Seamless card-based layouts for mobile devices, responsive tables for desktop
• ✓Secure Multi-User Authentication: JWT-based authentication with MongoDB for persistent data storage
• ✓Modular Architecture: Clean, maintainable codebase with separated component structure

Skills Highlighted:

Haver & Boecker

Enterprise Machinery Schedule Management System

A full-stack web application built for a real manufacturing client to manage machinery schedules, sales orders, and delivery milestones across two engineering departments.

Built on ASP.NET Core MVC with Entity Framework Core, this system manages complex relationships between orders, machines, vendors, and delivery milestones for Haver & Boecker's engineering and procurement teams. The backend handles multi-table relational data with foreign key constraints across sales orders, machines, vendors, and milestone tracking. An integrated Gantt chart built with the Frappe library visualizes engineering, procurement, and delivery timelines in real time. Led backend architecture decisions for the team, coordinated data model design across the full order lifecycle, reviewed teammate PRs, mentored on schema design, and maintained API endpoint documentation throughout development.

Technologies Used:

Key Features:

• ✓Real Client Delivery: Built for and deployed to an actual manufacturing client across two departments
• ✓Complex Relational Schema: Multi-table data model covering sales orders, machines, vendors, and delivery milestones
• ✓Gantt Chart Visualization: Real-time timeline view of engineering, procurement, and delivery phases using Frappe
• ✓Backend Architecture Lead: Owned data model design decisions and coordinated schema across the full team
• ✓RESTful API Layer: Clean API endpoints with full documentation maintained throughout development
• ✓PR Reviews & Mentorship: Reviewed teammate code, mentored on schema design decisions

Skills Highlighted: