AI/ML Engineer — building production AI systems, not demos.
Recent CS graduate who builds AI systems that ship to production. My internship work is live — a RAG pipeline handling 300+ daily healthcare assignments at 90%+ accuracy. I have a published paper in federated learning and a paper under review at NeurIPS 2026. I focus on the intersection of agentic AI, distributed systems, and MLOps.
Currently interested in: voice agents, agentic AI systems, RAG pipelines, production MLOps.
893 req/sec · p99 <170ms · 0% failure rate
Built from scratch on Redis — no Celery, no RQ. Priority queues via sorted sets, worker pool via Python multiprocessing, exponential backoff retry (3x), dead-letter queue, live FastAPI monitoring dashboard, Prometheus + Grafana instrumented.
Python Redis FastAPI Prometheus Grafana Docker Locust
90%+ accuracy · 300+ daily assignments · Live in production
RAG pipeline over patient records and staff availability. LangChain agent executes natural language scheduling commands — override assignments, reschedule visits, update patient data. Integrated end-to-end with FastAPI + PostgreSQL + Redis on AWS. Agentic CI/CD system via GitLab MCP with zero manual intervention.
LangChain FAISS FastAPI PostgreSQL Redis AWS Docker GitLab MCP
Published · IJSDR Vol 11 Issue 4 · Impact Factor 9.15
Federated learning framework with zk-SNARK cryptographic verification to prevent gradient poisoning. Byzantine fault-tolerant coordination — maintained liveness under 30% simulated node failure. gRPC reduced inter-node overhead 10% vs REST baseline.
Python PyTorch gRPC Federated Learning zk-SNARKs Blockchain
Multi-agent RAG pipeline with semantic retrieval (FAISS), LLM-based methodology critique agent, structured scoring agents, and citation graph traversal for automated related-work discovery.
Python LangChain FAISS FastAPI Multi-agent
LSTM + Multi-Head Attention hybrid for mid-price movement prediction on L2 order book data. Optimized inference to <10ms latency on tick-level data with 500ms prediction horizons.
Python PyTorch LSTM Multi-Head Attention Time Series
Benchmarked DARTS and evolutionary NAS strategies across 200+ architectures on CIFAR-10/NAS-Bench-201. Reproducible MLflow tracking across SGD, Momentum, and Adam optimizers.
Python PyTorch DARTS MLflow NAS-Bench-201
| Status | Title | Venue | Year |
|---|---|---|---|
| 🔄 Under Review | SCAR: SLO-Constrained Anytime Reasoning for Language Model Agents | NeurIPS 2026 | 2026 |
| ✅ Published | Veriblock-FL: A Blockchain-Based Verifiable Federated Learning Framework Using zk-SNARKs | IJSDR Vol 11 Issue 4 · IF 9.15 | 2026 |
| 📄 Preprint | NeuroChainOps — Privacy-Preserving Blockchain MLOps for Federated NAS | ResearchSquare | 2025 |
| Repo | What it is |
|---|---|
| PyroQueue | Distributed task queue — Redis, 893 req/sec, Prometheus |
| Veriblock-FL | Federated learning + zk-SNARKs — published paper |
| Graph_anamoly_detectio_GNN | Graph neural network for anomaly detection |
| Multivariate_Anomaly_LSTM_VAE | LSTM + VAE for time series anomaly detection |
| Active_Learning_Pipeline | Active learning pipeline for efficient labeling |
| Geospatial_Crime_Prediction | Geospatial ML for crime pattern prediction |
Open to AI/ML Engineer, AI Engineer, and SDE roles — Hyderabad, Bangalore, or Remote