Hello, I am

Jwalith Kristam

Full Stack AI Engineer & Data Scientist

Moving AI from correlation to causation. I build intelligent systems designed to shape a better future.

View Experience

About Me

Passionate about AI Innovation

3+ Years

Professional Experience in AI & ML

M.S. Data Science

Stony Brook University, New York

Technial Focus

GenAI, Classical ML & MLOps

I bridge the gap between theoretical research and production reality. With a strong academic foundation in statistical learning, I have spent the last 3+ years engineering AI systems that don't just predict outcomes but understand the why behind them. Currently, I am focused on creating autonomous Agentic Workflows and RAG systems that solve complex, real-world problems.

Generative AI & LLMs

PyTorch LangChain RAG LoRa Quantization vLLM Vector DBs

Data Science & Machine Learning

Causal Inference Bayesian Statistics CV NLP Supervised & Unsupervised

Engineering

Python SQL AWS Docker MLOps Redis Elasticsearch

Latest Articles

Thoughts & Insights

Text Diffusion

Exploring the capabilities of text diffusion models and their impact on generative AI. A deep dive into the architecture and potential applications.

Experience

My Professional Journey

Jul 2025 - Present

AI Engineer

Schizophrenia & Psychosis Action Alliance

Agentic RAG & Distributed Scraping Pipeline

GenAI RAG Distributed Systems Playwright Redis Geocoding Python

The Mission: Create a unified, searchable database of mental health housing resources across 50 states to help patients find care.

Data Engineering

Engineered a distributed scraping pipeline using Playwright, Celery, and Redis to normalize unstructured data from disparate state websites into a clean knowledge base.

GenAI Implementation

Built an Agentic RAG chatbot using LangChain and OpenAI to let users query this data naturally.

Optimization

Implemented Adaptive Chunking and MMR Re-ranking to improve retrieval precision by 40%, ensuring patients receive accurate, location-specific results without hallucinations.

UI & Geo-Search

Created an intuitive user interface for the resource search platform. Built a geo-search backend using Haversine distance with ZIP/city coordinates to locate nearby facilities. Implemented an incremental geocoding pipeline with OpenStreetMap API, cutting API calls by 90% and speeding data updates.

May 2024 - Aug 2024

Data Science Intern

Welspot

Predictive Financial Risk Engine & Strategy

AWS SageMaker XGBoost Optuna Causal Inference Tableau MLOps

The Mission: Transition the client from manual risk assessment to an automated, ML-driven financial risk engine.

Model Development

Benchmarked XGBoost against PyTorch RNNs on AWS SageMaker, deploying the XGBoost model to achieve 91% predictive accuracy with low latency.

Strategic R&D

Went beyond prediction by using Causal Inference and ANOVA to validate risk factors, identifying 7 key features that reduced false positives by 12%.

Business Intelligence

Created a reproducible MLOps workflow using SHAP values to explain model decisions, visualizing insights in Tableau for the executive team.

Oct 2020 - Jun 2023

Software Engineer

Tata Consultancy Services (Client: Silicon Valley Bank)

Real-Time Fraud Detection System (MLOps & Inference)

FastAPI Docker ONNX Runtime MLflow NumPy

The Challenge: The bank needed to detect fraud in real-time without slowing down transaction processing.

The Solution: Architected a high-throughput inference API using FastAPI and Docker.

Key Tech

Leveraged ONNX Runtime to achieve sub-50ms latency for 100+ concurrent requests.

Reliability

Engineered a custom drift detection framework using NumPy/SciPy to monitor PSI on live streams, integrating with MLflow to trigger automated retraining.

Customer CRM Automation (NLP)

Hugging Face BERT NLP PyTorch

The Challenge: Support teams were drowning in manual ticket reviews.

The Solution: Implemented an NLP-driven classification pipeline using Hugging Face transformers (BERT).

Impact

Automated routing for thousands of tickets, eliminating 2,500+ hours of manual work and saving $175K annually.

Banking Data Platform Modernization (Data Engineering)

Azure Synapse PySpark Elasticsearch Redis

Scale

Unified terabytes of fragmented data using Azure Synapse and PySpark ETL pipelines.

Performance

Optimized transaction history search using Elasticsearch and Redis, reducing query times by 40%.

Mar 2019 - Jun 2019

Machine Learning Engineer Intern

SmartBridge

IoT Pet Feeding Detection System

YOLO TensorFlow Computer Vision Kafka IoT

The Challenge: Pet owners needed a reliable way to monitor whether their pets were actually eating from automated feeders in real-time.

The Solution: Built and deployed a lightweight computer vision pipeline using YOLO and TensorFlow, connected to an IoT camera and streaming events through Kafka.

Model Engineering

Fine-tuned YOLO to detect pet presence and feeding actions under varying lighting conditions, achieving high detection accuracy on low-cost hardware.

System Design

Integrated the model into an edge-friendly pipeline with Kafka-based event streaming, enabling real-time alerts when feeding events were missed.

Open Source

Public Contributions

Optuna (GitHub)

View Repo

Refactored legacy Python code to modern Python 3.10+ standards by implementing f-string formatting in the visualization module.

Python Optuna Open Source Refactoring

Recommendations

LinkedIn Endorsements

“Kristam worked under my supervision designing and testing predictive ML credit risk models. I recommend Kristam for his commitment, drive, knowledge, and the contributions he brings to the table.”

Michael Montenegro Lederman Chief Strategy Risk Officer

View on LinkedIn

“I had the pleasure to hire and manage Jwalith during his internship at WelSpot. His knowledge of data, modeling, and machine learning was on point, and his approach to problem solving followed best practice.”

Tom John CIO | CTO

View on LinkedIn

Featured Projects

Innovation in Action

Voice-to-SQL: Real-Time Voice AI Data Analyst

A voice-activated data analyst that converts natural language speech into SQL queries using Gemini 2.5 Flash Lite.

TypeScript React gRPC Gemini AI WebSocket GraphQL

Adaptive Ad Recommendations

A hybrid system combining Reinforcement Learning for optimal strategy and LLMs for user context understanding to deliver personalized ad recommendations.

LinUCB RL LLM Deep Learning

Multimodal Sentiment Analysis

Deep learning system fusing text, audio, and visual data to detect nuanced emotional signals with high accuracy.

CLIP Multimodal Embeddings NLP Computer Vision

Music Recommendation System

Advanced recommendation engine using Two-Tower architecture and SASRec, achieving 85% Hit Rate@5 with GPU acceleration.

PyTorch CUDA Two-Tower FAISS

Historical Document Analysis with LLMs

Leverages LLMs and NLP to analyze historical apprenticeship agreements, extracting structured data (names, ages, locations) and uncovering socioeconomic patterns, geographical distributions, and temporal trends from unstructured historical text.

GPT-3.5 LLaMA LangChain NLP Pandas

YouTube Comment Analytics

Real-time streaming analytics pipeline using Kafka and PySpark to process YouTube comments with topic classification, toxicity detection, and engagement prediction using Hugging Face models.

Kafka PySpark Hugging Face BERT NLP

Mental Health AI Coach

An empathetic AI coach using DeepSeek-7B fine-tuned with LoRA and RAG. Features semantic retrieval, response reranking with Gemini 2.0, and encrypted session handling.

LoRA RAG DeepSeek Streamlit

Legion:Repair Assistant

I thought, "It would be great if I had an AI that could just look at the issue and tell me what to fix." So, I built Legion Repair Assistant.

Static Analysis Gemini RAG Voice

Quantized LLM Implementation

C++ implementation of quantization techniques for LLMs to optimize memory usage and inference speed while maintaining model quality.

C++ LLM Quantization Ollama

Deep Research Autonomous Agent

A multi-step reasoning agent capable of recursive web research. It plans, executes, scrapes, and synthesizes complex topics into comprehensive, citation-backed reports that mimic human research workflows.

LangGraph Agentic Tools Tavily API DeepSeek

Multi-Agent Travel Planner

A coordinated set of LLM agents that plan end-to-end trips, comparing routes, stays, and budgets across sources to produce optimized itineraries with rationale.

Multi-Agent Tool Use LLM Orchestration Python

Resume Job Matcher – Chrome AI Extension

A Chrome extension + companion website that analyzes resume–JD alignment, rewrites bullets, and generates tailored cover letters using Chrome's on-device AI APIs with a Gemini fallback for chat and rewriting.

Chrome AI APIs Prompt API Resume Matching Gemini

The Reading List

Books That Inspire Me

Siddhartha

by Hermann Hesse

A spiritual journey exploring self-discovery, enlightenment, and the path to inner peace.

A Psalm for the Wild-Built

by Becky Chambers

A Monk and Robot Book exploring purpose, connection, and what it means to be alive in a changing world.

Get In Touch

Let's Create Something Amazing

I'm currently open to new opportunities and collaborations. Whether you have a question or just want to say hi, I'll try my best to get back to you!