Skip to content
View prashanthjaganathan's full-sized avatar

Block or report prashanthjaganathan

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.

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

Hi, I'm Prashanth

ML engineer who ships AI systems end-to-end, from architecture and training through serving, evaluation, and production monitoring. I care about systems that hold up under real constraints, not demos.

MS in Artificial Intelligence, Northeastern University, Boston

Build carefully, instrument everything, stay accountable through delivery.


What I work on

I build production agentic and ML systems: LangGraph agents, multimodal RAG pipelines, LLM evaluation infrastructure, and the low-latency serving layers underneath them. Most of my recent work lives in healthcare and regulated environments, where correctness is not optional and the users are real people.

Recent highlights

  • First AI engineer at Polygon Health, where I shipped a conversational medical-records agent over patient FHIR data solo: guardrailed LLM-to-SQL, SSE streaming over a Postgres checkpointer, and request-scoped isolation. Cut p95 latency 38% and failed requests 91%.
  • Built an FDA regulatory document platform at GlaxoSmithKline from 0 to production in two months, with a multimodal RAG pipeline and an LLM-as-Judge eval framework in CI/CD that reduced hallucinations 42%.
  • Granted Indian Patent No. 575220 for an end-to-end waste management system spanning computer vision, IoT, and a solar-powered plastic-to-bio-oil pyrolysis reactor.

Selected projects

Boston Pulse — Civic intelligence platform for safe pedestrian navigation. GCP-native MLOps app with an Airflow-orchestrated pipeline training a LightGBM crime-risk model via GitHub Actions CI/CD, serving H3 safety scores from Firestore.

DRL Trading Agents — Rainbow DQN, PPO, and LSTM-Actor-Critic trading agents under a POMDP formulation. LSTM-PPO returned +49.3% on AMZN vs +27.9% buy-and-hold.

Medical Diagnosis Prediction — Clinical NLP models pretrained on 5M PubMed abstracts, achieving 20%+ improvement on MIMIC-IV diagnosis prediction through transfer learning.

ChoreMate — Full-stack agentic app using React, FastAPI, and Gemini to fairly assign and manage household chores with smart scheduling and swapping.


Toolbox

Python TypeScript · PyTorch LangGraph FastAPI React · Docker Kubernetes · GCP AWS · PostgreSQL Redis · RAG · LLM-as-Judge · Hybrid Search


Reach me

LinkedIn · jaganathanprashanth@gmail.com

Pinned Loading

  1. BostonPulse BostonPulse Public

    Forked from himabindu-peramala/Boston-pulse

    An ML-driven digital twin of Boston integrating civic, mobility, and public data.

    Jupyter Notebook

  2. SimulatingVirusSpread SimulatingVirusSpread Public

    Simulating the spread of COVID-19 in a largely populated workforce

    Python

  3. DRLAgents-for-Trading DRLAgents-for-Trading Public

    Forked from PurvajaNarayan/DRLAgents-for-Trading

    CS5180 Reinforcement Learning - Spring 2026

    Jupyter Notebook

  4. pretaining-language-models-for-medical-text pretaining-language-models-for-medical-text Public

    This repository focuses on pretraining language models on the MeDAL dataset for medical text, utilizing multiple architectures to enhance performance for medical NLP tasks.

    Jupyter Notebook 2

  5. choremate choremate Public

    Making household chores collaborative, fair, and actually enjoyable for shared living spaces

    JavaScript 1

  6. IamShrijan/GeneratedImageDetector IamShrijan/GeneratedImageDetector Public

    Model to classify as a geniuine or a AI detected Image

    Jupyter Notebook 1