Generative AI · Open source

RAG, LLM serving, and multi-cloud infrastructure.

I build retrieval systems and inference pipelines, then ship them on AWS, Azure, and GCP. I also send tested patches upstream — Python, Go, and TypeScript — instead of working around the bugs.

Gyanu Mayank

Gyanu Mayank

Generative AI Engineer

  • B.Tech CSE, NIT Patna · Pune
  • RAG, vector search, Terraform deploys
  • CodeChef 4★ · LeetCode Knight

About

Engineer first. Models are the current problem set.

I work on generative AI products: document retrieval, embedding pipelines, and getting models into production without the usual pile of glue. Competitive programming is how I got used to reading other people’s code and finding the actual bug.

NIT Patna, Computer Science. I contribute to libraries I depend on — parsers, servers, tracing, CLIs — with a regression test on each patch.

4★ CodeChef · peak 1917
1000+ LeetCode problems
8.25 CGPA · NIT Patna

Experience

Current work and the path into it.

Generative AI Engineer

Jun 2024 – Present · Pune
  • Multi-cloud AI pipelines on AWS, Azure, and GCP with Terraform and CI.
  • RAG stacks over vector databases, focused on retrieval quality rather than longer prompts.
  • Deploy and tune LLMs for lower-latency inference in production.

Bosscoder Academy

Content Creator Intern · Jan 2023 – May 2024 · Remote
  • Wrote 200+ pieces on DSA, ML, and core CS, with practice sets attached.
  • Ran coding contests and mock interviews.

Sanity Technologies

Software Developer Intern · Jan 2024 – May 2024 · Patna
  • Python REST APIs and database work for internal tools.
  • Debugging, docs, and server-side performance.

Trustedwear Tech

Software Developer Intern · May 2023 – Jul 2023 · Patna
  • Acoustic classification models above 80% accuracy.
  • Worked through Whisper model layers; about a 20% lift in prediction quality.

Earlier: problem setting & writing

TechCurators · iMocha · GeeksforGeeks · 2021 – 2024
  • Designed and reviewed 400+ DSA problems for hiring platforms (TechCurators, iMocha).
  • Wrote CS articles for GeeksforGeeks.

Open source

Merged patches in libraries other people ship.

Small, scoped fixes with a test. The names live in the table so the intro does not have to repeat them.

  • Chainlit CI fails when a dependency override is declared but never applied
  • Gunicorn Stop writing a 500 over a response that already started
  • django-extensions print-sql no longer crashes when the last query is missing
  • Logfire Keep traces until every span in them has finished
  • Python-Markdown Parse inline code by walking spans instead of a brittle regex
  • Waitress Don’t kill the event loop when a closed socket returns EBADF
  • environs Return None for invalid values when eager parsing is off

All merged pull requests →

Projects

Things you can open and run.

RAG · Python 2025

Advanced RAG

Document Q&A over PDF, DOCX, CSV and more — chunking, auth, Streamlit UI, Docker.

  • Multi-format ingest and retrieval, not a single-PDF demo.
  • Packaged for deploy (Dockerfile, compose).
LLM · Streamlit 2025

LinkedIn Post Generator

Few-shot post generation from a curated dataset, with Groq in the loop.

  • Layered Python app: UI, generation, and data kept apart.
  • Topic, language, and length as first-class controls.
Llama 2 2025

Llama 2 Chat

Streamed chat over 7B / 13B / 70B with tunable inference parameters.

  • Replicate-backed model switch without rewriting the UI.
  • Chat history kept in Streamlit session state.
React · Editor 2023

CloudCode

In-browser HTML / CSS / JS editor with live preview, built on CodeMirror.

  • Three panes, one preview — the usual sandbox loop.
  • Deployed on Vercel.
React · Firebase 2023

Document Editor

Docs-style editor: create and open documents, persisted in Firebase.

  • React Router for list vs editor views.
  • Documents stored remotely, not only in localStorage.
Python · Desktop 2023

Disk Analyzer

Tkinter tool for disk cleanup: duplicates, large files, temp files, rarely used files.

  • Duplicate detection by MD5 hash.
  • Desktop GUI, not a CLI dump.

Competitive programming

Ratings and a few contest results.

  • CodeChef — 4★, peak 1917
  • Codeforces — Specialist, peak 1455
  • LeetCode — Knight, 1000+ problems
  • Scaler HACKX 2021 — 44 / 15,500
  • Google Kick Start 2022 — rank 1264, Round H
  • TechGig Code Gladiators 2023 — finalist, 150k+ field

Education

National Institute of Technology, Patna

B.Tech in Computer Science & Engineering · CGPA 8.25 · Dec 2020 – May 2024

Python C++ TypeScript RAG AWS / Azure / GCP Terraform Docker React