AI Engineer · Data Scientist · Seattle, WA

Building the
future,
one model
at a time.

MS Data Science candidate at Seattle University. I build production-grade AI systems at the intersection of machine learning, hardware interfaces, and real human problems.

Scroll
Pavan Arram
Seattle, WA · 2024
Selected Work

Projects that ship.

01
Wearable AI · Hardware
JobHunt Glasses
Meta Ray-Ban Display API · Neural Band · Hands-free job apps in 60s
02
RAG · Production
FRAGAI
Hybrid BM25 + semantic RAG · 126 live users · $0.001/query
03
Agentic AI · Vision
Meal Prep AI
GPT-4o Vision · Amazon + Walmart APIs · 45s vs 20min manual
Scroll to explore
01
Wearable AI · Agentic Pipeline · Hardware

JobHunt
Glasses

Tap Apply on a job card displayed on Meta Ray-Ban Display glasses. 60 seconds later, a Claude-tailored LaTeX resume PDF lands in Gmail. Neural Band gesture → Vercel → Adzuna → Claude scoring + tailoring → tectonic LaTeX → Gmail SMTP. Hands-free. No laptop.

60s
End-to-end
600×600px
Waveguide display
6
Docker builds fixed
Meta Ray-Ban Display APINeural BandFastAPIClaude APItectonic LaTeXAdzuna APIGmail SMTPDockerVercel
Explore Project
↓ Keep scrolling
02
RAG · Production · Full-Stack

FRAGAI
RAG System

Production-deployed perfume recommendation RAG — hybrid BM25 + semantic search with Reciprocal Rank Fusion across 1,800 fragrances. Sub-500ms latency, $0.001 per query, 126 real user searches. CI/CD auto-deploys, 10 production issues resolved live.

<500ms
Query latency
$0.001
Per query
126+
Live searches
React + ViteFastAPIChromaDBBM25RRF (k=60)Claude HaikuDockerRailwayAWS S3
Explore Project
↓ Keep scrolling
03
Agentic AI · Computer Vision · Full-Stack

Meal Prep
Grocery AI

Paste a cooking reel URL. GPT-4o Vision samples keyframes + NLP parses captions. Custom ingredient resolver deduplicates, hits Amazon + Walmart APIs for live prices, forecasts 4-week grocery budget within ±8%, exports structured Excel. 20 minutes → 45 seconds.

92%
Extraction accuracy
±8%
Budget forecast
45s
vs 20min manual
GPT-4o VisionFastAPINext.jsSupabaseBullMQRainforest APIWalmart APIopenpyxl
Explore Project

More Work

Also built these.

🔬
Narrative Bias in LLMs
NLP Research

Quantified how GPT and Gemini distort moral framing, sentiment, and gender representation across 200 narratives. SBERT, BERTScore, BERTopic, UMAP, HDBSCAN.

SBERT 0.6042 vs 0.5769BERTScore F1 0.84-0.86
Hotel Demand Forecasting
Time-Series · BI

Revenue management forecasting across 3 NDA hotel properties. ARIMA, XGBoost, Random Forest benchmarked. Automated BI dashboards for stakeholders.

MAE 10.7-14.5MASE <0.530-day horizon
Distributed Task Queue
Distributed Systems · Azure

Async job processing with priority scheduling and dead letter recovery. Redis + Azure Service Bus. 94% message loss reduction, 99.2% uptime.

10K+ jobs/day99.2% uptime94% loss reduction
CascadeGuard
AI Safety · GPU Inference

Three-tier GPU-resident guardrail cascade with Bayesian confidence router. Fast embedding classifier under 5ms, LLM judge under 500ms.

<5ms tier-1Bayesian routingCUDA optimized
GPU Cluster Simulation Lab
MLOps · Kubernetes · CUDA

4 virtual GPU nodes simulating distributed AI training. DCGM + Prometheus + Grafana. Ansible + Terraform IaC cut cluster setup from 3 hours to 15 min.

4 GPU nodes25% throughput gain15min IaC setup
Skin Disease Diagnosis (IEEE)
Medical AI · Published

IEEE-published multimodal deep learning. Xception + ResNet-50 + LSTM late fusion on ISIC dataset (2,357 images, 7 classes). ~90% accuracy.

~90% accuracy7 disease classes3 citations

Experience

Companies I've built with.

Data Science Intern via Coincent
6 months · Azure Platform
Engineered and deployed a chatbot achieving 85% accuracy in query responses by applying advanced data processing and machine learning techniques. Earned Azure Fundamentals certification during the internship.
Azure Fundamentals AZ-900
Sports Performance Data Scientist
Jan 2026 – Present · Capstone
Hawkin Dynamics CMJ force plate analysis: PCA, KMeans athlete clustering by biomechanical profiles for coaching decisions.
Capstone Project
Operations Staff & Data Analyst
Jan 2025 – Present
MySQL + Streamlit platform: 12 tables, 20K+ records, dual BI dashboards for 60+ staff. 40% query improvement, 60% reporting reduction.
Live System

Publication

Peer-reviewed research.

📄
IEEE Xplore · DOI 10.1109/ICKECS61492.2024.10617139
Deep Learning-Based Predictive Skin Disease Diagnosis and Analysis
Manikundan Karnati; Pavan Sai; Rohith Kumar Pittala; K. Radhika; K. Sugamya
2024 ICKECS · April 18-19, 2024 · Chikkaballapur, India
View on IEEE Xplore →
3
Citations
189
Full-text views
~90%
Accuracy
Pavan Arram
Puget Sound, Seattle

About

Engineer,
researcher,
builder.

MS Data Science candidate at Seattle University, graduating June 2026. Applied ML, production AI systems, and emerging hardware interfaces.

From sub-$1 RAG pipelines to wearable AI co-pilots on Meta Ray-Ban glasses — I care about systems that work in the real world, not just notebooks.

IEEE-published. Azure certified. F-1 OPT STEM extension. No sponsorship required.

Machine Learning
LLM / Agentic AI
Production RAG
Wearable AI (BLE)
Azure Cloud (AZ-900)
Distributed Systems
Python · R · SQL
Sports Analytics

Get In Touch

Ready to build
something real?

Open to full-time roles in Data Science, ML Engineering, and AI Engineering.
Available June 2026 — no sponsorship required.