rawift

Forward Deployed Engineer

I’m a Founding Engineer at ProPeers, an edtech startup with 100,000+ users. I build AI products end to end, from the model layer through the backend to the UI. Recently I built Tracelyst, a content-protection platform that finds pirated media across Google, YouTube, Reddit and Telegram. It matches media using perceptual hashing, vector embeddings and an LLM judge, and a LangGraph agent with tool calling runs AI-driven scans. I’ve also shipped a multi-agent AI interview system, a RAG-based social media agent, and semantic search for 100k+ users. I refactored 150+ APIs to cut backend latency by 70%, and I run our Docker, Azure and CI/CD deployments with Prometheus and Grafana monitoring. 

My stack: Python, Node.js, TypeScript, Next.js, LangChain/LangGraph, OpenAI, vector DBs, MongoDB, PostgreSQL, Redis, Docker/Kubernetes, Azure and AWS. I’m a 2025 VIT CSE grad, a hackathon winner, and I’ve solved ~600 DSA problems. 

Looking for: remote full-time roles as an AI Engineer, Full-Stack Engineer or Founding Engineer at an early-stage startup. I want to work somewhere I can own AI features from prototype to production.


Experience: 2 years

Yearly salary: $30,000

Hourly rate: $60

Nationality: 🇮🇳 India

Residency: 🇮🇳 India


Experience

Founding Engineer
ProPeers
2024 - 2026
Architected an adaptive multi-agent AI interview system using OpenAI GPT-4, with dynamic question generation based on real-time candidate responses, an on-screen code editor with live execution, and AI-guided hints. Built an AI-powered social media agent using Python, LangChain and RAG, delivering real-time hashtag recommendations, competitor analysis and AI-generated SEO copy. Developed a learning roadmaps platform serving 100+ paths to 100,000+ users. Reduced API response time from 2.1s to 300ms (7x faster) with Redis caching, and cut Largest Contentful Paint from 4.4s to 990ms with Next.js SSR. Built a semantic search engine using cosine-similarity vector matching with Sentence-Transformers and custom OpenAI embeddings, powering natural-language queries for 100,000+ users. Refactored 150+ REST APIs with async execution, Redis caching and queue-based processing, reducing backend latency by 70%. Migrated MongoDB to self-hosted replica sets with automated backups. Implemented Prometheus and Grafana monitoring dashboards and managed zero-downtime deployments via Docker and GitLab CI/CD on Azure App Service.

Skills

ai
english