Varshneya Bhushan

Typeface.ai

• Established a framework-agnostic behaviour evaluation system for agents and skills with Langfuse tracing, specialized evaluators and ADO pipeline integration, grading each run on seven dimensions with LLM-as-judge and static graders, cutting manual agent testing by 80%.

• Owned production skill-based retail product-listing agents end to end for an enterprise CPG customer, cutting run latency 3x through measured prompt and context-compaction ablations and raising success rates from ~64% to 87-98% with a cron-based reliability harness, error taxonomy and CSV chunking strategies that return partial output on step failure.

• Architected Plan and Execute orchestration framework enabling multi-agent coordination for complex user requests, with support for human-in-the-loop interrupts, dynamic replanning with validation feedback and flexible iteration strategies for long-running LLM workflows.

• Built CSV bulk processing workflow achieving high reliability for thousands of records through optimized batching, retry logic, milestone tracking and structured output validation.

• Added integrations for Digital Asset Management platforms (e.g., OneDrive, Google Drive) to store assets generated using generative AI. Connected Email Marketing platforms (e.g., Mailchimp, Klaviyo) to automate email campaigns with AI-generated HTML content using Workato recipes.

Blueflagsecurity.com integrating connectors for various source control platforms like GitHub, Bitbucket to monitor and get insights about security risks.

• developed masterclass series to better organize learning material. Got 1 million views monthly, with 15 million minutes of watch time, with total of 1.3 million users

• developed a feature to save favourite material for students. improved retention rate by 40%

• made tools to write better scripts for automation, which is being used by 10 fellow developers, to do several tasks like fixing data inconsistency, data migration etc

• Built LangGraph-powered AI conversation engine with persona-based characters, automatic memory summarization, iterative quality refinement, Redis-backed state persistence, and multi-provider LLM support

• Architected notebook platform with AI character lifecycle management, turn-based orchestration, real-time WebSocket communication, dual-theme UI, and AWS S3-backed asset storage

• Implemented token-based billing system with usage tracking per interaction, provider-agnostic payment integration, multi-currency support, and real-time credit updates via WebSockets

• Engineered an AI resume builder using LangChain, incorporating an AI agent developed to help craft the resume, which analyzes job descriptions to automatically rewrite content for optimal matching, with automated PDF parsing and AI-generated cover letters based on profiles

• Built portfolio generation and hosting service (TypeScript, MongoDB, React 19) with multiple templates with OAuth authentication (GitHub/Google) and drag-and-drop UI.

an offline-capable JavaScript execution environment (React, IndexedDB, Service Workers) that runs entirely in the browser with zero server calls, allowing users to write, run, and persist code snippets even without internet connectivity.

a Python 2/3-compatible Sudoku solver package that applies constraint programming to solve generic n×n puzzles, returning structured outcomes including solution grid and solver metadata.