const tanush = {
role: "AI Platform Engineer",
education: "MS Computer Science @ UMass Amherst",
focus: [
"Production-grade RAG systems",
"Responsible AI and privacy safeguards",
"Full-stack AI platforms",
"Carbon-aware computing",
"Applied machine learning"
],
currentlyBuilding: "Reliable AI systems that move from prototype to production",
philosophy: "Ship useful systems. Make them fast. Make them trusted. Make them explainable."
};I build AI systems that make complex information easier to search, understand, and act on. My work sits at the intersection of LLMs, cloud infrastructure, full-stack engineering, applied ML, and responsible AI — from government AI assistants and privacy gateways to carbon forecasting platforms and multilingual document intelligence tools.
| 525+ State employees supported |
600+ Documents indexed |
9,000+ Webpages crawled |
<15 sec Search time achieved |
| 96% User satisfaction |
85+ Carbon regions covered |
340+ Data streams managed |
350+ Students supported |
|
Production-grade RAG assistant for state employees to search internal documents and web-crawled knowledge sources with cited answers. Built with: React · TypeScript · AWS CDK · AWS Bedrock · Claude · Docker · App Runner · Cognito What it did: Reduced manual document search from 10+ minutes to under 15 seconds while supporting secure authenticated workflows. |
Fail-closed PII/PHI redaction gateway for responsible AI adoption across public-sector chatbot workflows. Built with: Python Lambda · AWS Comprehend · Comprehend Medical · React · TypeScript · AWS CDK What it did: Masked sensitive entities before downstream model handling and helped make AI-assisted government workflows safer. |
|
Carbon intensity forecasting platform for visualizing and predicting low-carbon electricity windows across global regions. Built with: Django REST · React · TypeScript · Python · MapLibre GL · Time-Series ML What it did: Scaled carbon forecasting across 85+ regions and 340+ data streams with automated data recovery and region-specific predictions. |
AI-powered venture research assistant using hybrid search, reranking, and deterministic computation to reduce hallucination. Built with: TypeScript · React · Supabase · pgvector · BM25 · LLM APIs · Tailwind · Vitest What it did: Combined RAG, structured data workflows, and user-facing AI search into a practical startup intelligence platform. |
RAG Assistant Pipeline
flowchart LR
A[User Query] --> B[React + TypeScript UI]
B --> C[Cognito SSO]
C --> D[Containerized API]
D --> E[AWS Knowledge Base]
E --> F[Documents + Webpages]
F --> G[Claude via Bedrock]
G --> H[Cited Response]
H --> B
Privacy-Safe AI Gateway
flowchart LR
A[Raw User Input] --> B[PII/PHI Redaction Gateway]
B --> C[AWS Comprehend]
B --> D[AWS Comprehend Medical]
C --> E[Offset-Aware Masking]
D --> E
E --> F[Sanitized Prompt]
F --> G[Downstream Model Handler]
G --> H[Safe AI Response]
Carbon Forecasting Platform
flowchart LR
A[Regional Energy Data] --> B[Fault-Tolerant Ingestion]
B --> C[Feature Engineering]
C --> D[Forecasting Pipeline]
D --> E[Django REST API]
E --> F[Interactive Map UI]
F --> G[Researchers + Grid Operators]
AI / LLMs AWS Bedrock · Claude · LangChain · RAG · vector search · prompt engineering
Frontend React · Next.js · TypeScript · Tailwind · MapLibre GL · Leaflet · HTML/CSS
Backend Python · Django REST · Node.js · Express · REST APIs · WebSockets
Cloud AWS CDK · App Runner · Cognito · S3 · DynamoDB · Lambda · CloudWatch
ML / Data PyTorch · TensorFlow · pandas · NumPy · scikit-learn · LSTM · time-series ML
Dev Tools Docker · Git · GitHub · Vitest · pytest · Linux · Bash · Figma| Project | What it is | Stack | Focus |
|---|---|---|---|
| Commonwealth AI Search | RAG assistant for document and web search | React, TypeScript, AWS, Bedrock | Production AI |
| GENIE Privacy Gateway | PII/PHI redaction layer for chatbot workflows | Python, AWS Comprehend, CDK | Responsible AI |
| SLATE | Legal research chatbot for session laws | LLMs, streaming, cited retrieval | Legal AI |
| CarbonCast | Carbon intensity forecasting platform | Django, React, MapLibre GL, Python | Climate AI |
| VentureXify | Startup intelligence and RAG research assistant | TypeScript, Supabase, pgvector | AI Products |
| TextualPolyglot | Multilingual multimodal document intelligence | Python, LangChain, Gemini, pandas | Multimodal AI |
| Algo-Visualizer | Interactive algorithm learning tool | React, TypeScript, SCSS | CS Education |
| UPlay! | Campus sports connection platform | React, Node, MongoDB, Socket.io | Real-time Web |
-
LiteCast: A Lightweight Forecaster for Carbon Optimizations Lightweight carbon-intensity forecasting for real-time scheduling and carbon-aware computing.
-
ClaRet — A CNN Architecture for Optical Coherence Tomography CNN-based OCT scan classification system for retinal tear detection.
-
Visualizing the Impact of User Preferences and Habits on Carbon Emission Interactive visual analytics system for understanding carbon-aware behavior and energy usage patterns.
teaching:
role: Head Undergraduate Course Assistant
course: Python programming
students_supported: 350+ per semester
focus:
- labs
- grading
- office hours
- course logistics
- assignment and solution design
leadership:
role: President, Google Developer Student Club @ UMass Amherst
impact:
- revived inactive chapter
- led 7 officers
- organized technical workshops
- reached 600+ active participants
- raised funding for community technology access
01. AI systems with real users, real constraints, and measurable impact
02. RAG workflows that cite sources instead of hallucinating confidence
03. Privacy-first AI tools that handle sensitive information carefully
04. Full-stack platforms that make complex ML systems usable
05. Climate-aware tools that connect software engineering with sustainabilityI’m a builder, researcher, mentor, and sports obsessive. I like systems that are fast, thoughtful, and useful — whether that means a production AI assistant, a carbon forecasting map, a teaching tool, or a fantasy football model I will absolutely not make public.
Let’s build AI systems people can actually trust.