I'm a researcher at UMass BioNLP Lab working on multi-agent debate frameworks and RAG systems, and building at the intersection of applied AI, NLP systems, on-device ML, and agentic workflows. Former AI/Data Engineer at Deloitte with production experience in data platforms and ML infrastructure.
mindmap
root((Applied AI))
LLM Systems
Evaluation
Reasoning
Retrieval & Memory
RAG Pipelines
Knowledge Graphs
Agent Orchestration
Multi-Agent Systems
Debate & Reflection
ML Infrastructure
Production Deployment
Reliability & Explainability
🔍 What drives my work
- LLM systems and evaluation → Building reliable, measurable AI applications
- Retrieval, memory, and reasoning workflows → Making AI systems smarter and more contextual
- Graph + agent orchestration → Creating explainable, structured AI reasoning
- Production-grade data and ML infrastructure → Scaling AI from prototype to production
- AI products that work → Reliability, explainability, and real-world usefulness
|
|
An on-device memory assistant for dementia patients that remembers people, objects, and context using vision and voice — keeping all data local and private.
🔍 Why SnapOn Matters
Unlike cloud-based assistants, SnapOn runs entirely on-device using edge ML:
- ✅ Privacy-first: All data stays local on the phone
- ✅ Real-time: NPU-accelerated inference with Qualcomm Hexagon
- ✅ Offline-capable: No internet required for core features
- ✅ Production-ready: Built for Samsung Galaxy S25 Ultra
Tech Highlights:
- ExecuTorch 1.3 with Qualcomm QNN delegate for NPU acceleration
- MobileFaceNet for face recognition
- Whisper (on-device STT) + Piper (TTS)
- CameraX + ML Kit face detection
- Room Database for local memory storage
Use Cases:
- 📸 Remember where you put your keys or glasses
- 👤 Recognize family members and recall their relationship
- 🗣️ Voice-based memory retrieval and storage
- 🏠 Scene understanding for contextual assistance
My recent work has focused on turning AI ideas into usable systems:
|
Explores multi-agent debate, self-reflection, and training-free self-improvement in language models Key Features:
|
Multi-agent research assistant for paper retrieval and analysis Tech Stack:
|
|
Graph-native prototype for explainable patient follow-up reasoning Highlights:
|
Showcase of AI systems work and projects Built With:
|
📦 Detailed Tech Breakdown
Languages: Python, SQL, TypeScript, Java, C++
AI/ML: PyTorch, LangChain, LangGraph, RAG pipelines, LLM evaluation, NLP
Data & Infra: Spark, Databricks, Airflow, AWS, ETL/ELT, data pipelines
Apps & Tools: React, Next.js, Tailwind CSS, Streamlit, Neo4j, Git, Vercel
I’m currently interested in roles where I can work on:
Applied AI Engineering |
AI/ML Infrastructure |
LLM Products & Evaluation |
Data Platforms |
Technical Ownership |
I enjoy writing about AI systems, memory, reasoning, and practical implementation ideas alongside building them.
"I like building systems that don't just generate output, but reason better, retrieve better, and hold up better in real use."
