Skip to content
View ChandhanSaai's full-sized avatar

Block or report ChandhanSaai

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. Markdown is supported. 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
ChandhanSaai/README.md

Hi, I'm Chandhan 👋

🌐 chandhan.com: case studies with live demos, and the notes behind each project.

AI Engineer. I build LLM systems end to end — multi-agent pipelines, evals, RAG, and the production infrastructure they run on. Background in reinforcement learning and robotics, so I'm comfortable anywhere between "train the model" and "ship the system."

  • 🧠 LLM systems: multi-agent orchestration, adversarial evals, RAG, structured outputs
  • 🤖 Applied ML: transformers from scratch, reinforcement learning, computer vision
  • ⚙️ Production: Django services, AWS, Docker, Postgres, Prometheus — AI that actually ships

🔧 Featured projects

Project What it is
Project D3 — Multi-Agent LLM Evaluation Role-specialized LLM agents — advocates, a judge, and an optional jury — score model outputs through structured adversarial debate; two eval protocols with agreement, κ, and position-bias metrics.
Browser Autopilot A vision-driven browser agent: a LangGraph + Playwright loop that turns plain-English web tasks into actions and auto-generates illustrated step-by-step tutorials.
LLM-Triaged Security Audit Merges 8 SAST/secret/dependency scanners into one SARIF, then a two-pass LLM dedupes, filters false positives, and runs a semantic authz/IDOR reasoning pass into a ranked report.
GPT from Scratch A GPT-2-style transformer implemented from scratch in PyTorch — hand-built multi-head attention, weight tying, and a mixed-precision + gradient-accumulation training loop.
Reinforcement Learning — Panda Arm A Soft Actor-Critic agent (Stable-Baselines3 + PyTorch) trained in the robosuite simulator to control a Franka Panda arm on a manipulation task.
Autonomous Robot Perception — TurtleBot ROS 2 perception on a real TurtleBot: RANSAC vanishing-point detection, optical-flow obstacle stops, and a custom-trained YOLO detector driving lane-following in sim and on hardware.

🛠️ Tech I work with

LLMs & agents: multi-agent orchestration · RAG · evals / LLM-as-judge · prompt & context engineering · function calling · MCP ML: PyTorch · Transformers · Stable Baselines3 · reinforcement learning · YOLO · computer vision Languages: Python · C++ · TypeScript · SQL Production: Django · React · PostgreSQL · Docker · AWS · Prometheus


📫 Get in touch

LinkedIn · katurichandhansaai@gmail.com

Pinned Loading

  1. GPT-Transformer-Architecture-with-Production-Training-Pipeline GPT-Transformer-Architecture-with-Production-Training-Pipeline Public

    A complete implementation of GPT-2.5 transformer architecture from scratch with modern ML engineering optimizations, achieving production-level performance and scalability.

    Python 1

  2. Project_D3 Project_D3 Public

    D3 orchestrates role-specialized LLM agents -- advocates, a judge, and an optional jury -- to evaluate LLM outputs through structured adversarial debate, inspired by courtroom proceedings.

    Python 1

  3. multi-scanner-security-audit-with-llm-triage multi-scanner-security-audit-with-llm-triage Public

    One-shot, full-codebase security audit pipeline: 8 SARIF scanners merged and LLM-triaged (dedupe, FP filtering, exploitability ranking, semantic authz/IDOR pass). Runs ephemerally on EC2 and self-t…

    Shell 1

  4. Metered_API_Billing Metered_API_Billing Public

    A metered API billing system in Django + React: usage ingestion, hourly aggregation, monthly tiered invoicing, mock payment webhooks, and customer + ops dashboards. Built for correctness under conc…

    Python

  5. -Reinforcement-Learning-Based-Pick-and-Place-Task-for-Panda-Robot -Reinforcement-Learning-Based-Pick-and-Place-Task-for-Panda-Robot Public

    This project leverages robosuite, a simulation framework for robotic manipulation, in combination with Stable Baselines3 (SB3) to train and evaluate a Soft Actor-Critic (SAC) agent for the Pick-and…

    Python 7

  6. ROS_2_TurtleBot_perception_project ROS_2_TurtleBot_perception_project Public

    ROS_2_TurtleBot_perception_project

    Python 1 2