Data Science & Machine Learning

Data Analytics & Visualization

Data Engineering & Infrastructure

Cloud & Full Stack

Development Tools & Automation

Sustainability & Environmental

🗓️ Oct 2025 – Present   |   📍 Tempe, AZ

• Client-Facing IoT Platform: Architected an IoT water-monitoring platform for 9 stations, 14 sensor parameters, and 1.4M+ records at 5K+ readings/day — built directly for Salt River Project (SRP) as a client, now in active production use by SRP and 10+ ASU researchers.
• Ingestion Reliability: Engineered checkpoint recovery and idempotent 500-record batch ingestion achieving 99.9997% ingestion success, catching a ~960K-row desync incident via daily reconciliation.
• Performance Engineering: Migrated time-series storage to TimescaleDB and layered Redis caching, cutting full-fleet retrieval from 1.3-3.9 hours to under 2 seconds and improving API latency up to 57x.
• Anomaly Research: Ran a 14-round benchmarking process across sklearn IsolationForest and RandomForestClassifier models, exposing sensor-data contamination and reaching 0.80 F1 / 0.85 recall on 7,352 held-out windows.

🗓️ Feb 2025 – Oct 2025   |   📍 Tempe, AZ

• Geospatial ELT: Automated an ELT pipeline integrating CWNS 2022 and NHDPlus federal datasets, cutting data processing time 60% with schema validation and bad-record rejection.
• Graph Optimization: Independently optimized memoized BFS traversal over a 1.07M-node river DAG for 1,100 intake points, cutting traversal time from 95 seconds to 12 seconds (8x speedup).
• Predictive Analytics: Modeled flow imputation with sklearn GradientBoosting for 337,000 ungauged river reaches, achieving 0.87 R², with features consumed by 3 research groups.

🗓️ May 2024 – Dec 2024   |   📍 India

• Retention Modeling: Modeled logistic regression and XGBoost on 500K+ user activity records to predict early learner churn, improving retention prediction 18% over rule-based baselines.
• Database Optimization: Migrated PostgreSQL to AWS RDS, indexed hot columns, and cut p99 query latency 40%, while schema validation reduced ingestion errors 80%.
• AI Support Assistant: Deployed a LangChain + OpenAI embeddings RAG assistant with Chroma/FAISS retrieval, reducing dashboard-tracked support tickets by about 30%.

🗓️ May 2023 – Aug 2023   |   📍 India

• Campaign Optimization: Designed 2 Wix web variants and ran an A/B test across 3K+ visitors, measuring a 15% relative CTR lift to inform UX decisions.
• Referral Growth: Co-developed a 12-ambassador referral campaign, analyzing 8,500 referral visits and measuring a 4.3% conversion rate — about a 15% relative lift over non-ambassador traffic.
• Reporting Automation: Automated weekly SQL and Python reporting, cutting turnaround from 3-4 hours to about 1 hour, backed by 2 Tableau dashboards tracking 8 KPIs.

AI-native food-truck marketplace spanning 107 trucks and 635 menu items. A 4-tier stack (Next.js, Spring Boot gateway, FastAPI/LangGraph AI service) powers an AI Concierge, a structured-output Order Builder, and an Owner Copilot for sales and complaint analysis.

Focus: Agentic AI · Full-Stack AI Product · Tool-Calling
Tech: Next.js · Spring Boot · FastAPI · LangChain/LangGraph

Agentic medical RAG platform indexing 9K+ medical chunks with FAISS and cross-encoder reranking, plus a 3-step ReAct trace for auditable answers. Benchmarked across 100 prompts and 5 medical domains for grounded accuracy.

Focus: Medical RAG · Agentic AI · Healthcare
Tech: ChromaDB · FAISS · LangChain · Next.js

Fashion computer-vision and hybrid recommendation engine combining CLIP embeddings with cold-start-aware weighting. Reaches 0.73 Precision@10 and 89% catalog coverage, with GPT-4o-mini explaining every recommendation.

Focus: Computer Vision · Recommender Systems
Tech: PyTorch · CLIP · FastAPI · Next.js

AIOps anomaly detector polling Kubernetes pod memory every second, using an LSTM + LightGBM hybrid to forecast failures before they happen. A custom Go Prometheus exporter feeds predictions into the autoscaler, with LLM-generated remediation commands.

Focus: AIOps · Kubernetes · SRE
Tech: TensorFlow · LightGBM · Go · Prometheus

Accessible speech platform built for stuttered, accented, and fragmented speech, reaching 87% intent-recognition accuracy. Real-time lip-synced 3D avatars map audio to phoneme cues at 95% match precision.

Focus: Accessibility · Speech AI · Real-Time 3D
Tech: React Three Fiber · Ollama · ElevenLabs

Retail demand forecasting across 1,115 stores. BigQuery ML ARIMA_PLUS models generate 33,450 30-day forecasts, benchmarked at 16.92 MAPE versus 21.91–24.74 for baseline models, delivered through a FastAPI backend and Streamlit dashboard.

Focus: Time-Series Forecasting · Data Science
Tech: BigQuery ML · FastAPI · Streamlit · Python

Deep Learning Specialization
DeepLearning.AI

Optimization for real-world AI systems