Anthony Maio
AI Product Engineer | Agents, Harnesses & Applied AI | 20 Years Scrappy Engineering
Professional Summary
Staff engineer with 15+ years delivering production systems across AI, financial services, and identity infrastructure. Currently leads architecture and delivery of the Native Agentic Engine at Pieces, spanning tool selection, context compaction, model evaluation, and large language model (LLM) cost accounting; directs implementation through coding agents, owning engineering decisions, diagnosis, review, and verification. Track record of measurable production impact: 99.99% login uptime, ~20% auth latency improvement, ~$654K/year vendor and infrastructure savings, and event-sourced systems sustaining ~5K tx/sec at <10ms latency. Combines research-grade prototyping (model training, evaluation harnesses, preference optimization) with production shipping discipline (CI-gated validation, rollout gates, incident response, observability).
Core Skills
Professional Experience
AI Staff Engineer
Apr 2026 – PresentPieces · Remote
Leads architecture and delivery of the Native Agentic Engine, progressing from productionizing the Rust/Dart/Flutter harness to context compaction and tool selection. Directs implementation through coding agents, owning engineering decisions, diagnosis, review, and verification; delivered ten consecutive native releases across Windows, macOS, and Linux.
- •Designed and integrated RouteAdvisor, a planner-based tool selector that gives agents task-specific toolsets and ordered execution guidance; added relevance-ranked memory retrieval for targeted recall.
- •Improved recovery from oversized memory results through context-relative ingestion limits and truncation; added turn-aware history retention and incremental summaries for long sessions.
- •Diagnosed prompt-cache and cost-accounting defects involving unstable tool ordering, cache-token normalization, and uncapped native tool results; corrected Gemini cache anchoring through OpenRouter.
- •Traced a Windows startup crash to leaked compiler flags through release-DLL disassembly (AES-NI instructions unsupported by affected CPUs); fixed artifact builds to target baseline x86-64.
- •Informed automatic model-tier selection through paired model/reasoning benchmarks and production-history replay, comparing task scores, tool use, latency, cost, and failures.
- •Built and shipped Pieces Video Factory, adopted by the team for product demos; combined AI scene planning, footage generated with ComfyUI, and editable local composition with reusable media.
- •Engaged Pieces users on Discord and X, promoted the product through a weekly AI newsletter, and contributed to the company's technical blog.
- •Authored architecture RFCs on execution persistence, retrieval-based tool selection, recovery from planning misses, and cross-session learning.
Staff Artificial Intelligence Engineer
Sep 2024 – Apr 2026Making-Minds.ai · Remote
Independent AI platform engineering practice delivering LLM infrastructure, agent reliability tooling, and AI safety evaluation systems. Clients include C-level stakeholders at startups and mid-sized industrial organizations (manufacturing/IoT, mineral processing). All artifacts published as open-source with reproducible training/evaluation configurations.
- •Designed and built agent execution and verification workflows treating agents as production actors with explicit latency/cost budgets, deterministic replay/verification gates, and CI-gated trust scoring for tool outputs and generated code.
- •Implemented sandboxed, headless tool-execution infrastructure with constrained permissions, auditable failure modes, and documented operational tradeoffs for agent orchestration deployments (Agent Hardening Pack).
- •Shipped LLM evaluation and monitoring tooling focused on regression detection and failure classification under operational constraints (throughput/latency/cost), with automated CI pipeline integration for model/prompt/tool change validation.
- •Trained and released the Eve-2 model family: a 272M-parameter Mixture-of-Experts base model pretrained from scratch on ~10.5B tokens (FineWeb-edu) using PyTorch DDP, plus instruction-tuned and task-specialist derivatives optimized for CPU/edge inference.
- •Built Argos-Swarm, an automated red/blue teaming and orchestration framework for multi-model adversarial safety evaluation, including cross-model epistemic divergence measurement for detecting weak-verified failures in LLM outputs.
- •Applied preference optimization (DPO/RLHF-style) to fine-tune GLM-4.7 for domain-specific protocol adherence; documented quantization/pruning tradeoffs under controlled adversarial testing (PV-EAT framework).
- •Delivered AI product engineering reviews and modernization roadmaps to C-level stakeholders, covering inference infrastructure architecture, model lifecycle management, AI platform governance, and cost optimization strategies.
Staff Software Engineer, Identity & Access Management Platform
Jan 2023 – Aug 2024DraftKings · Remote
Technical lead and architect for an 8-engineer IAM platform team supporting authentication, authorization, and compliance infrastructure across DraftKings' consumer and enterprise product lines.
- •Led and mentored 8-engineer remote IAM platform team; materially improved delivery throughput through tightened ownership norms, structured review workflows, paired programming practices, and pragmatic process standardization.
- •Executed zero-downtime 2FA/SMS provider migration: maintained 99.99% login uptime, improved auth latency ~20%, and reduced annual vendor spend by ~$450K; established organization-wide MFA standards across all platforms.
- •Reduced AWS infrastructure spend by ~$204K/year through right-sizing, targeted serverless adoption, Kubernetes HPA autoscaling tuning, and resource utilization optimization.
- •Architected and operated Kafka-based compliance data pipelines (Kafka to Redshift), processing ~5TB/day to support regulatory reporting, audit requirements, and KYC/AML compliance workflows.
- •Shipped AI-assisted development tooling (code review automation + knowledge retrieval workflows) adopted across 7 engineering teams; reduced review cycle time ~75% and measurably unblocked feature delivery velocity.
Staff Software Engineer, Trading Systems (New Ventures)
Aug 2022 – Dec 2022DraftKings · Remote
- •Delivered 0-to-1 MVP for a greenfield crypto trading platform; drove technical strategy and architecture decisions across two distributed engineering teams during early-stage product development.
- •Built event-sourced CQRS core sustaining ~5K transactions/sec at <10ms p99 latency (PostgreSQL + Kafka), establishing the reliability and performance foundation for the trading platform.
- •Designed and implemented self-service experimentation infrastructure that reduced feature launch cycle time from days to <1 hour, enabling rapid product iteration and data-driven decision-making.
Staff Product Engineer, Enterprise FX Platform
Jan 2015 – Aug 2022Broadridge Financial Solutions · Remote
- •Led staged modernization from monolithic .NET/WCF architecture to AWS microservices for an enterprise FX platform (UBS partnership; ~$2B AUM context), delivering incremental MVPs while onboarding pilot enterprise clients.
- •Architected Kafka-based orchestration spanning ~22 interdependent business processes across 7 financial institutions; designed for consistency/partition tolerance tradeoffs with full operational visibility and monitoring.
- •Enabled material revenue impact through reliability improvements, incremental platform adoption, and staged delivery methodology that de-risked enterprise client migrations.
Senior Product Engineer, Trading & Risk Platforms
Jul 2007 – Dec 2014TwoFour Systems (acquired by Broadridge Jan 2015) · Remote
- •Founding engineer building trading and risk platforms for Tier-1 financial institutions in client-facing, on-site delivery roles; delivered real-time risk/margin capabilities as a key product differentiator.
- •Designed and shipped production risk calculation engines processing real-time market data feeds, supporting portfolio-level margin and exposure monitoring for institutional trading operations.
Education
Bachelor of Science, Computer Science, May 2007, 3.74 GPA
Binghamton University — Thomas J. Watson College of Engineering and Applied Science, Binghamton NY