Home
About Me
I'm pioneering AI-driven enterprise automation that transforms how large organizations operate. Currently at AT&T, I'm building systems that learn from human experts and autonomously execute complex workflows across mainframe and web systems turning what once took hours into automated processes that complete in minutes.
My work centers on multi-agent AI systems powered by LangGraph orchestration, Retrieval-Augmented Generation (RAG), and vector databases. I design platforms where AI doesn't just assist but learns, adapts, and acts autonomously. When a domain expert records a workflow, my systems analyze it, generate semantic embeddings, and store it as reusable intelligence. When similar scenarios emerge, the AI retrieves and executes the optimal workflow without human intervention creating a continuously evolving automation network.
With deep technical expertise in production ML systems, GPU optimization, and distributed computing combined with strategic thinking from my business foundation, I build scalable AI solutions that deliver measurable enterprise impact. I thrive at the intersection of cutting-edge research and real-world deployment, converting decades of procedural knowledge into intelligent, self-improving systems that scale human expertise infinitely.
Education
-
Master of Science in Computer Science, University of Texas at Arlington
- Key Courses: Advanced Machine Learning, Data Structures & Algorithms, Big Data, AI, Cloud Computing, Neural Networks.
-
Graduate Diploma in Deep Learning, University of Texas at Arlington
- Highlights: Neural Networks, Computer Vision, Data Analysis & Modeling Techniques.
- B.Tech in Computer Science and Engineering, REVA University
Professional Experience
-
GenAI Engineer, AT&T
- Engineered AT&T's Agentic Foundry (AAW 2.0) — a LangGraph-based multi-agent orchestration framework automating 85+ enterprise workflows with decoupled reasoning, dynamic tool discovery, and autonomous error remediation
- Architected modular Python microservices with multi-tenant isolation, enabling domain experts to author natural-language workflows without code intervention while maintaining strict enterprise boundaries
- Built end-to-end agent pipeline (Retrieval, Triage/Planner, Validation, Task Runner, Remediation, Summarization) with Golden Workflow learning, reducing manual design overhead and elevating system reliability through reusable automation blueprints
-
Automation Engineer, Microsoft
- Architected event-driven multi-agent framework on AKS with LangGraph + Azure Functions, achieving 99.92% availability
- Deployed GPU model serving with Triton + vLLM, delivering ~2× throughput and 35–50% lower cost/1k tokens
- Built hybrid retrieval system reducing unsupported answers by 30%, citation pass rate improved from 69% → 92%
-
AI Engineer, Optum
- Built HIPAA/GDPR-compliant drug-discovery platform unifying PDB/ChEMBL/PubChem, TCGA/GEO, ClinicalTrials.gov
- Accelerated target discovery by ~20% using AlphaFold-style structural features with genomic Transformers
- Improved trial recruitment with GraphSAGE/GAT and BioBERT/Med7, achieving sub-second updates via Kafka + Flink
-
Machine Learning Engineer, Quora
- Shipped production recommender stack with two-stage ANN retrieval (FAISS/Pinecone) and learned ranking
- Built real-time features at scale using Kafka/Flink pipelines with online/offline feature store parity
- Optimized for multiple objectives (CTR, read time, quality, fairness) using multi-task learning
Skills
Publications and Research Papers
-
BudgetMem: Learning Selective Memory Policies for Cost-Efficient Long-Context Processing in Language Models
- Achieved 72% Memory Savings with Near-Perfect Performance – Developed BudgetMem, a novel memory-augmented AI system that reduces memory usage by 72.4% while maintaining 99% accuracy (only 1% F1 score drop) when processing long documents, enabling resource-constrained devices to handle book-length texts that previously required expensive, high-end hardware
- Democratizing AI: Running Advanced Language Models on $10/Month Hardware – Created a selective memory architecture that makes sophisticated long-context AI accessible to researchers and developers without expensive GPU infrastructure, validated on Google Colab to prove that intelligent memory management can replace brute-force computational power
-
Low-Level GPU Profiling & Kernel Debugging
- Developed systematic profiling playbook using nvprof, Nsight Systems/Compute, and NCCL traces to pinpoint bottlenecks in warp divergence, uncoalesced memory, and launch overhead
- Delivered kernel-level improvements (kernel fusion, shared-mem tiling, async copy, CUDA Graphs) achieving double-digit latency reductions and significant throughput gains on inference workloads
- Packaged optimizations as CI performance gates and production runbooks for deployment teams
-
Automation Pipelines
- Architected event-driven multi-agent pipeline with sagas, retries, and idempotent checkpoints reducing turnaround from hours to minutes
- Cut SME prep time by ~60% for document generation and evaluation workflows
- Implemented quality at scale with offline eval harness (pass@k, citation correctness, hallucination rate) and guardrails (RAG attribution/NLI checks)
- Raised citation pass rate and lowered unsupported outputs on held-out corpora
Key Projects
-
Custom Transformer Training Pipeline
- Developed transformer-based language model entirely from scratch using PyTorch
- Built end-to-end training infrastructure without relying on pre-trained weights
- Utilized distributed data parallelism (DDP) to scale training across multiple GPUs
- Designed custom data preprocessing pipelines and model architecture from the ground up
-
GPU Kernel Profiling & Optimization for Deep Learning
- Focused on low-level systems work to optimize deep learning inference by diagnosing and refining CUDA kernels
- Used profiling tools such as nvprof, Nsight, and NCCL to analyze GPU performance
- Implemented custom CUDA kernel enhancements resulting in up to 25% reduction in inference latency
- Leveraged kernel debugging techniques to streamline performance bottlenecks
Blog Posts
Coming soon! Check back for insights on AI/ML engineering, GPU optimization, and production ML systems.