Kushagra Bansal

Node.js, Fastapi, Python, React Native, GCP, Supabase (PostgreSQL), Supermemory, LiveKit, OpenRouter, Vectors (Jan 2026 - Present) • Building 0 to 1, Antler Backed early phase startup. • Developed real time voice AI companion using LiveKit + streaming LLM responses (+ Text chat with SOcket.io) • Built multi-agent LLM workflows with OpenRouter for matchmaking, voice interactions & real time text and voice AI chat • Creating lang graph for complex orchestrations & researching on knowledge graphs to pivot from vector embeddings. • Built long term memory pipelines using embeddings, Supermemory and Redis layers (for storing contexts), reduced response latency by 60% via caching and optimised retrieval, projected to improve session retention by 20–35%. • Designed tool-calling workflows allowing agents to dynamically retrieve user preferences and match data, with potential to improve match relevance and match CTR by 35% and user retention by 15% with improved realtime voice chat response. • Created orchestration loops using Temporal for multi-step agent workflows: match recommendations, evals, memory updates (user preference & persona), context analysis, image embedding generation etc, replacing ad-hoc queue/cron pipelines and reducing workflow failure recovery overhead, lowering operational costs by 25% through automatic retries & state persistence. • Designed horizontally scalable AI system architecture (Redis caching, async workflow orchestration, streaming inference, and vector retrieval) to handle 150K+ projected monthly users in the second quarter of the year. • Improved agent goal completion rates and session duration through optimized prompt chunking and dynamic context ingetion. Improved semantic retrieval efficiency by over 30% by compacting the vectors using Dimensionality Reduction.

Node.js, MongoDb, AWS, Lambda, GraphQL, REST, Stripe, Javascript, Docker, Redis(July 2024 - January 2026) • Designed and developed a complete server-driven solution for Stripe terminal integration using AWS services like SAM, Lambda, and AppSync. Enabled POS payments on physical devices by using GraphQL APIs and AWS Cognito for secure authentication. • Implemented webhooks to create a gateway for 3rd party services & GraphQL subscriptions to create a channel for publishing backend events that frontend can capture. • Optimised codebase with AWS Lambda Layers, significantly reducing code repetition, did error handling for 250+ apis

nodejs, redis, elasticsearch, mongodb, aws, firebase, pm2, remote ssh tunneling, cassandra, mysql, caching(June 2023 - July 2024) • Designed efficient architecture for, and implemented Instagram like stories feature with complete privacy management. • Implemented Automatic User Ranking System similar to competitive games (Catered 30,000+ DAU). • Poll system (Voting) with dynamic poll creation and realtime user engagement using Websockets. • HLS Microservice for videos using EFS, Python, Redis, SQS and Custom Webhook. • Delivered interest based videos, platform trending videos 1st iteration working model in just 3 days. • Modularized recommendation config, populated in Redis to update settings with API, eliminating backend release • Designed & implemented another video recommendation algorithm using Matrix Factorization. • Reduced feed retrieval apis response times by 50% & reduced server operations by changing elasticsearch indexing for posts • Optimised database performance by implementing lazy delete technique instead of realtime. • Migrated cloud infra from AWS to GCP with Google SADA team. Automated data migration b/w MongoDB, Elasticsearch, Redis. Developed Python scripts to automate complex tasks, streamlining processes & minimizing manual intervention.

React.js, Next.js, Vue.js, Nuxt.js, Redux, Tailwind CSS, Bootstrap, GraphQL, MUI, TypeScript(May 2023 - June 2023) • Edumim - Built responsive we app UI elements with proper state management and animations like carousels, accordion, sidebar, navbar with parallax, testimonials, sitewide dark mode, rich text, etc. • GSL & PE - Combined Nuxt2 & Nuxt3 projects and aligned tailwind configs, modules & components for compatibility. • WShipt & Xhibiter - Combined Next12 & Next13 projects and aligned tailwind configs, modules & components for compatibility.

Native android, Java, RESTFUL APIs, Google volley, Picasso, Glide, Gson, SQLite, Visualization(April 2022 - October 2022) • Built complete healthcare & ecommerce app with various health trackers, calculators, monitors, order tracking, coupons. • SSO for Google, FB & phone number and email sign-in using JWT token auth, Razorpay and PayU payment gateways. • Step tracker with real-time location with Google maps API, weight, sleep, period trackers, etc.