Skip to content
View kushagra0526's full-sized avatar

Block or report kushagra0526

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
kushagra0526/README.md

Hi, I'm Kushagra Pratap Singh

Typing SVG


About Me

I'm a final-year B.Tech student in Communication and Computer Engineering at LNMIIT, Jaipur (graduating 2027). I build backends and AI systems, and I care about the part that comes after the demo: tests, evals, and what happens when the input is messy.

  • I build LLM agents with tool calling, memory, approval steps and evals, mostly in Python with LangGraph.
  • I write backends in Node.js and Python: microservices, GraphQL gateways, Kafka, role-based access and CI/CD.
  • I was an SDE intern at Marine Edge Technologies, where I built role-based access and Razorpay payments for a platform with 7,000+ users.
  • I research multi-objective optimization with evolutionary algorithms under a professor, with a paper in mind.
  • I have solved 450+ DSA problems on LeetCode and GeeksforGeeks.

I'm currently looking for internships in backend and AI engineering. The best way to reach me is email or LinkedIn.


Featured Projects

An agent for support and operations work. It reads orders, a stock feed and policy documents, builds a plan, checks the plan against a deterministic risk policy, stops for human approval on risky steps, and logs every action for audit. It re-plans when a step fails.

I tested it on 12 scenarios and 8 held-out cases that I wrote separately. The held-out set found three bugs in my own code, which I fixed.

Python LangGraph Groq ChromaDB SQLite Streamlit


A gym support assistant with four agents (triage, booking, credits, cancellation) that hand customers to each other through tool calls. A Pipecat voice interface runs the same graph with speech-to-text and text-to-speech. An eval harness scores routing, latency and hallucination, and every turn is traced in LangSmith.

Python LangGraph Pipecat FastAPI Claude LangSmith


Upload a research paper and ask it a question. Every answer is a verbatim span from the document, shown with its page number, so you can check it yourself. I fine-tuned BERT to 85.05% F1 on SQuAD 2.0, retrieve passages with TF-IDF and re-rank them with a cross-encoder, and serve INT8 ONNX models through FastAPI inside a 512MB Docker container. I measured each stage: scoring takes about 29 ms and re-ranking adds about 730 ms.

Python PyTorch BERT ONNX FastAPI React Docker


A task planner where Gemini turns freeform text into structured fields and a dashboard compares the AI's time estimate, your own estimate and the actual time. PostgreSQL with Prisma and a pgvector HNSW index over 1536-dimension embeddings. Stats are aggregated server-side, so no raw task data reaches the LLM.

Node.js PostgreSQL Prisma pgvector Gemini React Recharts


Three independent services (user, product, order) behind an Apollo GraphQL gateway, with Kafka for events between services, JWT roles for buyers, sellers and admins, Joi validation, Swagger docs, Jest and Supertest tests, Docker, and CI/CD on GitHub Actions. A Next.js storefront sits on top.

Node.js Express GraphQL Apollo Kafka MongoDB Docker Next.js


A collaborative code editor backend with a CRDT (RGA) that I wrote from scratch, with no Yjs or Automerge, so edits from several people typing at once converge on the same document. Every operation goes into an append-only MongoDB log, with JWT-secured Socket.io rooms.

Node.js Express Socket.io MongoDB CRDT


Earlier projects

  • VoiceNotebook-QA: speak a note, store it locally and search it later. TypeScript, Vite, React, Web Speech API.
  • Stock Predictor: an ANN that predicts Apple's next-day price from OHLC data, with Random Forest feature selection.
  • AgriSys: a seed bank management system for small farmers in India. JavaScript, Node.js, MongoDB.

Tech Stack

Languages

Backend and APIs

Frontend

Databases

AI and ML

DevOps and Cloud


Achievements

🏆 Runner-Up, HackCrux Hackathon (Oarfin, an AI disaster-guidance platform)
🧩 450+ DSA problems solved on LeetCode and GeeksforGeeks
🎓 GenAI Workshop certificate at LNMIIT, scored 100/100
👥 Lead a 7 to 10 member Creative Team at the ACM LNMIIT Student Chapter
🤝 Sponsorship Head of Vivacity, LNMIIT's annual fest

Reach Me

Pinned Loading

  1. ecom_microservice_graphql ecom_microservice_graphql Public

    A scalable e-commerce backend built with a microservices architecture. Features a GraphQL API gateway, three independent services for users, products, and orders, async communication via Apache Kaf…

    JavaScript 2

  2. Interlace Interlace Public

    A real-time collaborative code editor built on a from-scratch CRDT (RGA algorithm) — no Yjs, no Automerge. Handles concurrent edits, offline reconnection with op queuing, JWT room auth, and rate li…

    JavaScript

  3. Later-Probably Later-Probably Public

    Later, Probably tracks the gap between predicted and actual task outcomes to surface procrastination patterns. Built with Express, PostgreSQL, Prisma, and JWT auth. Gemini powers freeform task extr…

    JavaScript

  4. moo-gui-platform moo-gui-platform Public

    A modular web-based platform for multi-objective and many-objective optimization built on top of the Pymoo framework. It provides interactive problem modeling, customizable evolutionary algorithms …

    Python

  5. PaperQA PaperQA Public

    PDF question-answering system. Extracts text from research papers, retrieves the most relevant passages using sentence-embedding similarity, then runs a fine-tuned BERT model to pull the exact answ…

    Python

  6. OpsPilot OpsPilot Public

    Autonomous support-operations agent that answers requests from internal knowledge (vector DB, SQLite, CSV, JSON), plans multi-step actions, pauses for human approval on critical ones, recovers from…

    Python