Senior AI/ML Engineer · Technical Lead · Meta AI
Building agentic LLM systems at Meta that process 10M+ posts daily. Specializing in post-training optimization, inference at scale, and ML systems that ship to billions.
01 — About
Senior AI/ML engineer and technical lead at Meta AI, driving post-training and inference optimization for multi-modal LLM systems at massive scale. Former founder of Momento AI, and engineering alumnus of AWS, Coinbase, TikTok, and Magic Leap.
My work lives at the boundary of systems engineering and machine learning — designing agentic AI pipelines, hardening them against adversarial prompt injection and tool misuse, and squeezing maximum throughput out of GPU inference infrastructure for production workloads serving billions of users.
Pursuing an M.S. in CS (ML/AI) at Georgia Tech while shipping production systems — because the best engineers never stop learning.
02 — Experience
Technical lead for agentic AI infrastructure processing 10M+ posts daily. Orchestrated post-training inference optimization pipelines spanning semantic search, vector-based document retrieval, and multi-modal image & video compression. Architected defensive layers against prompt injection, tool misuse, and data exfiltration in agentic LLM workflows. Focused on memory efficiency, latency reduction, and real-time system-level performance analysis in high-throughput GPU environments.
Founded and scaled a hybrid multi-cloud mapping and migration platform powered by open-source GenAI on a GPT-like interface. Built the full product from zero — managed a cross-functional team of PMs, UX designers, and engineers, growing the core team to 5. Full architecture ownership across Docker, Kubernetes, and Terraform spanning AWS, GCP, and Azure with SDKs in Python, Go, and Java.
Led design of fraudulent detection within in-house microservices as part of AI-driven payment initiatives. Achieved 75% latency optimization for payment transaction pagination serving global customers. Mentored a new hire and returning intern on technical skills and team best practices.
Reduced cost of customer nodes (Participation, Query & Transact) by 40–60% for on-chain validators. Researched blockchain infrastructure across Zero Knowledge (ZKML), Bitcoin forks, and Tendermint protocol.
Designed a data-driven system that increased external partnership throughput 5× to 50–100K QPS. Contributor and mentor on the identity graph and personalized ads recommendation engine at global scale.
Hardened backend services against OWASP Top 10 vulnerabilities. Designed a computer vision model using RCNN to detect and redact faces on Magic Leap One AR devices. Built secure shared cloud services to scale heavy spatial AR computations.
03 — Projects
Interactive systems map of how foundation models are trained, adapted and served—from tokenization and Transformer internals to MoE routing, KV caching, continuous batching and distributed inference. Exploded 3D atlas with lifecycle modes, a component inspector and twelve working microscope simulations.
Deep learning visual search engine using k-means clustering and visual vocabulary correlations. Built a bag-of-words representation over CNN feature embeddings to enable content-based image retrieval at scale.
Personal web platform and full-stack UI project maintained as a git submodule within the portfolio codebase. Explores modular front-end architecture and component-driven design patterns.
Parallel M/M/1 queue server simulation of CSMA/CA. Models wireless network behavior under variable load to analyze throughput, latency, and collision rates using queuing theory.
Custom programming language with minimal syntax and functional programming semantics — built from scratch with its own lexer, parser, and interpreter. An exploration of PL theory and compiler design.
Kaggle ML competition — trained an ensemble predictive model using gradient boosting and logistic regression to classify credit loan risk. Feature engineering, cross-validation, and hyperparameter tuning on real financial datasets.
App interface for poker admin and cashier tasks including game scenarios and chip management. Full UI for real-time game state tracking and session management for live poker games.
04 — Skills
05 — Research & Advising
Deep learning ensemble model for accurate volumetric segmentation of brain tumors using the BraTS dataset — combining multiple architectures to improve precision across tumor sub-regions.
Founding Partner and AI Advisor. Directed technical strategy for a platform using digital twins and health agents for early endometriosis diagnosis. Advised leadership on applied LLM, RL, and medical ML.
Built generative AI systems and infrastructure for Biotech companies using Transformers, NLP, and early LLM architectures — one of the earliest production applications of large language models in life sciences.
06 — Contact
Open to senior engineering and staff roles, AI infrastructure opportunities, research collaborations, and advisory conversations. Bay Area based — let's connect.