Open to Opportunities

I build AI systems
that ship — and scale.

From 0-to-1 LLM agents at JerseySTEM to a Chrome extension with 60K views in 48 hours. Production agentic AI, knowledge graphs, and systems that don't break.

See My Work arrow_downward Let's Talk
0

Views in 48 Hours

Taylor AI — Zero Marketing

0

Ops Overhead Cut

JerseySTEM Agents

0

RAG Precision

Up from 58% Baseline

0

Model Compression

Efficient Local Fine-Tuning

Scroll expand_more
school Rutgers M.S. Data Science
article IEEE Published Researcher
emoji_events 2x Hackathon Winner
deployed_code Production LLM Systems
hub Knowledge Graphs & Graph RAG

The Journey

From Research to Production AI

Every system I've built taught me something the last one couldn't. Here's how I got here.

rocket_launch
Aug 2025 — Present Current

Applied AI Engineer

JerseySTEM, NJ

  • Designed and shipped vertical AI agents using LangGraph and PydanticAI with deterministic tool routing and fault-tolerant DAG execution — reducing manual operational overhead by 65% across Google Workspace and Jira workflows.
  • Built schema-aware RAG pipeline on AWS Bedrock with hybrid retrieval, Redis context caching, and Cohere Rerank; improved precision from 58% to 84% with hallucination and retrieval-miss failure-mode tracking.
  • Delivered NL-to-SQL analytics engine with schema grounding and structured output validation; containerized on AWS Lambda + FastAPI, cutting data-to-insight turnaround by 40%.
  • Stood up production LLM observability using LangSmith and OpenTelemetry — full workflow tracing, TTFT monitoring, and prompt regression alerts — raising system reliability by 35%.
LangGraph PydanticAI AWS Bedrock Cohere Rerank FastAPI LangSmith
emoji_events
2026 1st Place

Rasa Boston Hackathon

Most Resilient Long-Term Agent. Built an agentic system that maintained coherent, goal-directed behavior across extended multi-turn interactions.

View Project arrow_forward
science
Nov 2023 — May 2025

AI/ML Engineer & Research Assistant

Rutgers University, NJ

  • Fine-tuned LLaMA-3 using LoRA/QLoRA on domain corpora; 4-bit quantization trimmed GPU memory by 38% with 97% accuracy retention; built reproducible eval harness for model safety and regression testing.
  • Built a decoder-only Transformer from scratch — custom multi-head attention, causal masking, and positional encoding — for distributed training and model safety research.
  • Deployed FastAPI + Docker inference services handling 15,000+ daily requests at sub-200ms p95 latency with zero unplanned downtime across 12 months.
  • Automated Airflow ETL pipelines with version-controlled snapshots and schema drift detection.
LLaMA-3 LoRA/QLoRA PyTorch Docker Airflow
emoji_events
Oct 2024 1st Place

HackRU Fall 2024 — NJ Transit Track

XGBoost delay prediction (500K+ records, 82% accuracy) + schema-aware RAG chatbot serving 15K+ queries. 70% faster FAQ resolution.

View on Devpost open_in_new
school
Sep 2023 — May 2025 GPA: 3.65

M.S. Data Science

Rutgers University, New Brunswick

Focus: LLMs, Generative AI, MLOps, System Design

biotech
Aug 2021 — May 2023

AI/ML Research Assistant

VNR VJIET, India

  • Applied 8-bit quantization to mBERT and IndicBERT, cutting memory footprint 40% with negligible accuracy loss across 5 multilingual NLP benchmarks.
  • Constructed Python + PostgreSQL ETL pipelines ingesting 2M+ unstructured documents with automated drift alerting, sustaining 99.5% operational uptime.
mBERT IndicBERT PostgreSQL Quantization
article
2023 IEEE Publication

Multi-Classification of Respiratory Diseases using Deep Learning

2023 ICSCSS — DOI: 10.1109/ICSCSS57650.2023.10169597

school
Aug 2019 — May 2023 GPA: 3.3

B.Tech Electronics & Communication

VNR VJIET, India

Specialization: AI & Machine Learning

flight
Oct 2024

Data Science Virtual Intern

British Airways via Forage

NLP sentiment analysis on 3,949+ reviews; built predictive models to map buying patterns and proposed experiments for ~20% campaign ROI lift.

What I've Built

Featured Projects

Systems that shipped, scaled, and solved real problems.

Generative AI NLP RAG

Taylor: AI Career Stylist

Problem

Job seekers spend hours tailoring resumes and cover letters for each application.

Solution

RAG-powered Chrome extension that auto-generates tailored career documents from any job listing.

Impact

60,000 views and 600 access requests in 48 hours — zero marketing spend.

Add to Chrome open_in_new Code
auto_awesome
60K+

Views in 48 hrs

600+
Requests
$0
Marketing
On-Device ML iOS Accessibility

GemmA.I: Vision Assistant

Problem

Blind and low-vision users lack real-time, private scene understanding tools.

Solution

iOS app with real-time scene understanding via MediaPipe + Gemini. 100% on-device privacy.

Impact

Sub-500ms inference with INT8 quantization. Fully private, zero cloud dependency.

Watch Demo play_arrow Code
visibility
<500ms

Inference Latency

INT8
Quantized
100%
On-Device
Agentic AI Rasa Stateful

Wally: The Always-On AI Coworker

Concept

Moves beyond the ephemeral 5-minute chat window to become a long-term, stateful digital coworker.

Features

Persistent state across sessions, proactive anomaly detection (heartbeat monitor), agentic triage with human-in-the-loop, and strict deterministic fallback flows.

Award

1st Place — Most Resilient Long-Term Agent (Rasa Boston Hackathon 2026).

View Code
robot_2
Always-On

Persistent State AI

Proactive
Heartbeat Monitor
Agentic
Triage & MCP Tools
Fine-Tuning memory

BetterLlama

Efficient locally fine-tuned LLaMA 3.2 for personal agent intelligence. LoRA + 8-bit quantization achieving 4x compression, 1.8x faster inference, 60% cost reduction. Integrated chain-of-thought and LLM Guardrails.

View Project arrow_forward
Transformers sports_esports

ChessGPT

GPT-based chess move prediction with custom tokenizer generating ~30 legal moves per prompt. Compact model footprint demonstrating Transformer architecture mastery.

View Project arrow_forward
XGBoost Agents
train

NJ Transit Assistant

1st Place HackRU. XGBoost delay prediction (500K+ records, 82% accuracy) + schema-aware RAG chatbot. 70% faster FAQ resolution.

Devpost open_in_new

Technical Depth

Competency Map

Core competencies across the AI development lifecycle — from research to production.

psychology

AI & ML

LLMs Agentic AI RAG Generative AI NLP Transformers Fine-Tuning LoRA / QLoRA Quantization Knowledge Graphs Graph RAG Prompt Engineering
build

Frameworks

PyTorch LangChain LangGraph PydanticAI HuggingFace TensorFlow Scikit-learn XGBoost FastAPI Rasa
cloud

Infrastructure

MLOps Docker AWS (Bedrock, Lambda) GCP CI/CD Harness Vector DBs PostgreSQL MongoDB Redis Airflow
code

Languages

Python SQL Swift JavaScript TypeScript Bash

Ready to build
something real?

Currently open to AI engineering roles and research collaborations. I don't bring demos — I bring systems that ship.

mail valluruchetanreddy@gmail.com
LinkedIn GitHub