Chidhambararajan Ramachidambaram

Software Engineer - Data Scientist

Read my technical blogs at blogs.chidha.dev

I am a person who loves applied research. I enjoy keeping up with new research breakthroughs and look for ways to incorporate them into solving day-to-day problems. That is one of the main reasons behind choosing AI and CSE as my career path. I hope to extend that same passion in your organization and grow along with you.

Work Experience

JPMorgan Chase & Co.

Software Engineer II (Data Scientist II) | Full-time

  • Joined as a full-time employee on a new team producing traffic forecasts for consumer-facing products.
  • Contributed 14+ forecasting models in four months, owning model development and continuous deployment.
  • Played a key role in migrating the MLOps pipeline from an internal cloud to Databricks.
  • Transformed the model-integration workflow, reducing onboarding time by 90% by removing redundant deployment failures.

Stack: Airflow, Docker, Prophet, NeuralProphet, CI/CD with Jules, AWS, Internal Cloud, Databricks

Contract Software Engineer - AI | Full-time contract role

  • Led a customer-support chatbot that resolved close to 50% of first-category tickets in its MVP.
  • Developed a self-healing Text-to-GraphQL engine with AST validation and runtime feedback, ensuring 100% schema compliance.
  • Delivered the GraphQL system to production as a firm-level product, available to authorized employees.

Stack: PyTorch, LangChain, LangGraph, Docker, AWS, Lambda Functions, CodeDeploy

Trademarkia

Lead AI Engineer | Full-time

  • Led the company's venture into AI and created an AI-powered logo search handling 12M+ images, scaled through SageMaker inference endpoints.
  • Created the Mark Registrability Scorer, which predicts the likelihood of a new trademark application's registration and serves as the landing point for the flagship Trademarkia.ai website.
  • Worked on a legal chatbot based on RAG techniques; led and mentored AI interns on architecture, code, and MLOps decisions.
  • Owned and implemented cost-effective MLOps pipelines for multiple AI projects.

Stack: PyTorch, Docker, Scikit-learn, LangChain, AWS, Lambda Functions, SageMaker

EDGENeural.ai

AI Developer | MLOps

  • Built PyTorch implementations of popular AI models and created training, optimization, and deployment workflows for the company's AI-as-a-service platform.
  • Added 18+ models, from classification to segmentation, to the platform.
  • Helped develop the company's proprietary automated document-information extractor, used by a client for portfolio-statement analysis.

AI Intern

First step into enterprise-level AI.

BG Enterprises Ltd.

Computer Vision Developer | Part-time

Created AI models that run on low-powered devices for:

  • FarmGuard: Animal intrusion prevention system for farmlands, which received local-government funding.
  • FaceMark: Face-recognition-based attendance and automated door-lock system.

Education

B.Tech. in Computer Science

VIT University, Amaravati

Selected Projects

FrontierScout

Automated Agentic Framework for Deep Learning Architecture Search

  • Architected an open-source, Optuna-inspired orchestration library that automates iterative deep-learning architecture and loss-function research through LLM-powered research agents.
  • Implemented an autonomous loop with rolling memory, automated code execution and retry harnesses, and multi-metric Pareto-frontier optimization.
  • Designed a modular tool-calling layer for live web research and engineered a typed, scalable Python package using uv and clear configuration specifications, preparing it for PyPI distribution.

github.com/TheSeriousProgrammer/FrontierScout

Replicated Microsoft BitNet in One Week

LLMs with 1.58-bit precision

Recreated BitNet's quantization approach within one week to validate its 1.58-bits-per-weight claims.

Read the write-up: BitNet 1.58-bit using PyTorch from scratch

EfficientWord-Net

Few-shot learning-based hotword detector

  • Built an open-source few-shot hotword detector, reducing conventional data requirements of 100K+ samples.
  • Project has received 310+ stars, 11+ community pull requests, and 20+ forks.
  • Ported the core inference pipeline to TypeScript/JavaScript for 100% client-side, real-time microphone processing.

github.com/Ant-Brain/EfficientWord-Net

Secure-Pass

End-to-end encrypted password manager inspired by Bitwarden, built with Flask, MongoDB, Flutter, and Heroku for a cloud-computing assignment.

Yoga Pose Detection

Classified multiple Surya Namaskar poses using few-shot learning.

Publications

EfficientWord-Net: Few-shot learning-based open-source hotword detection engine

World Scientific | Journal of Information & Knowledge Management

Skills

Computer VisionCUDADeep LearningElasticsearchLinux DesktopMachine LearningMLOpsPythonPyTorchRaspberry PiTensorFlowTensorRT