willastley.dev

Resume

AI Agent Architect focused on LLM runtimes, multi-agent orchestration, and production AI infrastructure.

William Astley
AI Agent Architect
wrastley@gmail.com
(518) 902-5805
Toronto, ON, Canada
US & Canadian Citizen

Summary

Agent systems engineer specializing in LLM runtimes, multi-agent orchestration, and production AI infrastructure. Creator of Haive, an open-source framework for adaptive multi-agent systems that move beyond fixed workflows by enabling agents to discover capabilities, compose tools, spawn sub-agents, and coordinate through stateful serializable graphs. Built 100+ agent implementations and 30+ open-source frameworks spanning agentic AI, quantitative systems, and developer tooling.

Professional Experience

Creator
Haive Framework
2024 - Present

Architected Haive's runtime and standardized ACP-style deployment services for dynamic multi-agent systems, with stateful serializable graphs, persistent memory, runtime reconfiguration, capability discovery, and MCP integrations.

  • Built composable agent-team orchestration with planners, reasoners, memory, dynamic supervision, parallel execution, and validation and repair loops for adaptive workflows.
  • Developed configurable RAG and context pipelines across 200+ sources, spanning custom ingestion, document processing, embedding, indexing and storage, retrieval orchestration, and 20+ RAG patterns.
  • Designed evaluation and benchmarking systems across 22 game and application environments, using fixed-config ablations to compare tools, prompts, models, and agent architectures.
Senior Quantitative Risk Developer
Corpay, Toronto
2022 - Present

Led Corpay's agentic AI initiative by building multi-agent systems on Haive, including a KYC and AML watchlist assistant that achieved 96% detection accuracy, 5x faster throughput, and full coverage over sampling.

  • Built a database intelligence agent and credit underwriting assistant for internal teams, improving data access and enabling faster, more consistent credit reviews.
  • Engineered an automated forecasting and stress-testing platform using GARCH-based models for VaR, volatility, and scenario analysis, improving accuracy by 35% and reducing runtime by 75% across $500M.
  • Built and maintained a centralized multi-repo platform, unifying CI/CD across credit, treasury, and trading.
Founder & Lead Developer
Algebraic Wealth Technologies, Toronto
2022 - 2023

Built and led a 10-person team developing EquityExpert, an early multi-agent platform for automated sell-side equity research with live, data-driven reports spanning fundamental, technical, macro, and quantitative analysis.

  • Advanced to the Y Combinator interview stage for AI-driven analytics and document-ingestion systems.

Technical Skills

Agentic Systems & Evaluation
LangChain
LangGraph
LangGraph SDK
LangSmith
Deep Agents
MCP
Agent/runtime design
Graph composition
Multi-agent orchestration
A2A protocols
Persistent memory
RAG
Knowledge graphs
Prompt design
Validation
HITL
Benchmarking
Agent observability
Deployment
Languages & Backend
Python
SQL
TypeScript
FastAPI
Pydantic
SQLAlchemy
Pandas
NumPy
AI/ML & Quantitative Methods
PyTorch
TensorFlow
Hugging Face Transformers
Qdrant
pgvector
Stochastic modeling
Time-series analysis
Statistical evaluation
Systems, Infrastructure & Observability
PostgreSQL
Docker
Redis
MinIO/S3
OpenTelemetry
Prometheus
Grafana
Loki
Jaeger
Temporal
Celery
RabbitMQ
Developer Tooling & CI/CD
Git/GitHub
GitHub Actions
Trunk
pre-commit
Commitizen
pytest
coverage
reporting
Ruff
Mypy
Pyright
uv
PDM
Poetry
Hatch
Nox
Sphinx
Build and release pipelines
Versioning

Selected Open-Source Platforms & Projects

Full portfolio, documentation, and package releases via willastley.dev , GitHub and PyPI.

Agentic AI Platform
haive-core

Stateful graph runtime with serializable agents, tools, retrievers, and engines embedded in layered state schemas for reconfigurable, persistent, and reproducible execution.

haive-agents

Advanced orchestration layer for composing planner and reasoner teams with parallel DAG execution, validation and repair flows, MCTS-style reasoning, context management, and knowledge-graph and vector memory.

haive-tools & haive-mcp

Ecosystem of 100+ standardized tools and MCP integrations, enabling agents to discover, retrieve, and extend capabilities with HITL-gated access across 2,000+ tools.

haive-games

Agent and multi-agent benchmarking suite spanning 22 environments with fixed-config ablations over models, prompts, and architectures to capture reasoning, coordination, and robustness.

haive-dataflow & haive-hap

Deployment framework and ACP/MCP-style protocol serving agent graph workflows with JSON-RPC execution, lifecycle hooks, recovery, tracking, and Postgres-backed persistence.

haive-prebuilt

50+ configurable agent presets and templates for common enterprise workflows, including support, research, content generation, code review, analytics, and domain-specific automation.

Agent Runtime & Developer Infrastructure
perplexity-at-home

Open-source Perplexity-style research engine with quick search, pro search, and deep research implemented as LangGraph workflows for planning, retrieval, reflection, and synthesis.

sqldbagent

Database intelligence agent using DB-RAG and long-term memory for schema-aware exploration, guarded query planning and execution, and iterative self-improvement.

oas2mcp

Agentic OpenAPI-to-MCP system that transforms API specs into refined FastMCP and LangGraph server artifacts, improving tool docs and generating prompts, resources, templates, and server instructions.

ooai-promptdb

Database-backed prompt registry and runtime delivery layer for LangChain and LangGraph with relational versioning, file-native prompt specs, and Postgres and MinIO-backed storage.

ooai-skills

Multi-agent skills registry that ingests curated skills repositories, normalizes SKILL.md capabilities, and enables portable install, sync, and discovery across Codex, Claude Code, Cursor, and agent runtimes.

ooai-llm

Unified LLM integration layer for LangChain-first applications with provider-aware model parsing, typed settings, live model discovery, LiteLLM metadata, and cost helpers.

pjkm

Agent-native project generation framework for creating structured, reproducible coding environments with project context, tooling, commands, and conventions encoded up front.

Agent-Logic

Pydantic-based agent toolkit for structured, validated reasoning with logic ASTs, sequent-calculus proof validation, SAT checks, machine-checkable traces, and counterexamples.

Prompt-injections

Defensive prompt-injection guide and corpus with 200+ adversarial cases, attack taxonomy, detection rules, mapped defenses, and few-shot and evaluation integrations.

notex-ai

Early agent-style pipeline for converting handwritten notes, PDFs, and images into structured LaTeX through OpenCV preprocessing, OCR, LLM generation, and self-correction.

Quantitative AI / Financial Systems
wraquant

Unified quantitative and market-systems platform spanning time-series forecasting, stochastic and PDE simulation, technical analysis with 265 indicators, derivatives pricing, risk modeling, stress testing, backtesting, and agent-facing financial workflows.

Credit-Default

Credit risk modeling pipeline for default probability estimation using logistic regression, KNN, and XGBoost with feature engineering and evaluation via ROC-AUC, Brier score, and calibration.

Education

Theoretical Mathematics Specialist Program Honors BSc
University of Toronto
2018 - 2021

Recruited into industry prior to degree completion based on technical achievement and open-source contributions.