Jay Gupta.
I build and own production backend systems and AI-driven products with correctness, cost-awareness, and long-term impact.
AI Engineer and builder. I ship production-grade systems fast.
About
I'm a Computer Science (Data Science) undergraduate with hands-on experience owning backend systems and AI-heavy features in production. My work centers on async APIs, ETL pipelines, AI agents, retrieval systems, and automation workflows. I care deeply about system correctness, operational cost, and engineering tradeoffs — especially when building AI-powered products meant to scale beyond demos and hype.
Experience
AI Engineer
Dec 2024 – Jan 2026Edtech Startup with 100k+ learners
- →Owned a daily end-to-end ETL system combining web scraping, normalization, AI summarization, and structured storage using cron-based automation. Directly contributed to a 3× increase in app downloads and a 60% increase in daily active users.
- →Architected and shipped a production-grade GRE AI Tutor using Pydantic-AI, with schema-validated outputs, tool-based reasoning, persistent user context, and performance tracking.
- →Built the complete async backend for GRE Quant, Verbal, and Vocabulary modules using FastAPI, async SQLAlchemy, asyncpg, and PostgreSQL, focusing on predictable latency and clean domain separation.
- →Engineered subscription and access-control infrastructure using Razorpay, including free trials, international payments, secure webhook verification, idempotency handling, and real-time entitlement checks.
- →Reduced AI infrastructure costs by approximately 50% through batch inference, prompt restructuring, token optimization, and deliberate model selection based on cost–quality tradeoffs.
- →Built scalable OCR and document-ingestion pipelines using GOT OCR 2.0 and Gemini 2.5 Pro to digitize and structure physical and scanned educational content.
- →Developed internal automation systems including AI-assisted video generation workflows using n8n and custom MoviePy pipelines, significantly reducing manual content operations.
- →Prototyped early-stage realtime voice-to-voice agent for interactive use cases, exploring latency, streaming responses, and conversational state management.
Selected Work
SketchPenAI Whiteboard Video Generator
Script in, voiceover added, animated whiteboard video out in minutes. Custom rendering engine built from scratch. Full API access at every pricing tier. $0.70 per standard video. Solo-built in 7 effective days across 21 strict iterations, feature parity with a YC-funded competitor.
ApexAI Desktop Assistant
Initial Versions of NotebookLM frustrated me. no chat history, no streaming, bad memory, hard-to-read fonts. So I built Apex: a Cursor-like desktop AI assistant with persistent local notes, document uploads, font customization, type-safe agents, Rust-powered filesystem service, and full MCP support. The AI environment I actually wanted to use.
Technical Focus
Backend & Systems
- FastAPI, SQLAlchemy
- PostgreSQL, asyncpg, Alembic
- Docker, Nginx, Linux
- AWS, Cron-based Automation
AI & Retrieval Systems
- Pydantic-AI, LangChain, LangGraph
- MCP, MultiAgents, Agent Skills
- RAG Pipelines
- TTS, OCR, Voice Assistants
- Transformers, PyTorch
Languages
- PythonPrimary
- TypeScript
- Rust
- SQL
Client & Desktop
- React, Tailwind CSS
- Tauri v2Native Apps
- Next.js