Post Job Free
Sign in

Software Engineer

Location:
Henrico, VA
Posted:
November 17, 2026

Contact this candidate

Resume:

WILLIAM SCOTT COCHRAN

Richmond, VA 804-***-**** *****.*******@*****.***

linkedin.com/in/scott-cochran-b069624 github.com/scottcochran PROFESSIONAL SUMMARY

Senior Software Engineer and Systems Architect with 25 years of experience in carrier-grade and bank-grade distributed systems, event-driven architecture, and cloud-native infrastructure. Proven track record designing secure, scalable, fault-tolerant systems: high-throughput Apache Kafka data pipelines, resilient backend services in Python and Node.js, and fine-grained policy enforcement with Open Policy Agent (OPA). Delivered the first production Docker platform and the first production Kafka pipeline at Capital One, and co-invented a patented geo-redundant streaming platform. Author of TOPAK, an open-source, formally verified architecture for governed agentic AI systems.

TECHNICAL SKILLS

Languages and Runtimes: Python, Rust, Node.js, Go (Golang), OPA Rego, Cedar, TLA+, Lean 4, SQL, GraphQL, Bash

Distributed Systems and Messaging: Apache Kafka, Kafka Connect, Multi-Paxos Consensus, Event-Driven Architecture, Microservices, REST APIs, Write-Ahead Logs Cloud and Infrastructure: AWS (ECS, EKS, Lambda, S3, RDS, Aurora, EFS, DynamoDB, SageMaker), Docker, Kubernetes, Mesos, CI/CD Pipelines, OpenTelemetry, New Relic Big Data and ETL: Hadoop, HDFS, Spark, Apache Flume, Apache NiFi, Parquet, Avro, JSON, Snowflake Data and Policy Governance: PostgreSQL, NoSQL, DynamoDB Single-Table Design, Data Modeling, Open Policy Agent (OPA), Cedar, IAM, SSO, ABAC, RBAC, Policy as Code, Signed Policy Bundles AI and Agentic Systems: Multi-Agent Orchestration, Model Context Protocol (MCP), Tool Calling, Agentic Workflows, Policy Guardrails, Prompt Injection Containment, Human-in-the-Loop Architecture, Retrieval- Augmented Generation (RAG), Vector Databases (ChromaDB, Milvus), LLM Evaluation Formal Methods: TLA+ Specification, TLC Model Checking, TLAPS and Lean 4 Proofs, Refinement, Trace Replay

Architecture and Leadership: System Design, Architecture Decision Records (ADRs), High Availability, Disaster Recovery, Security, Compliance, Technical Leadership, Cross-Functional Teams SELECTED PROJECT

TOPAK - Safe-by-Design Architecture for Agentic AI Open Source 2025 - Present github.com/scottcochran/TOPAK Apache-2.0 220+ commits 55 Architecture Decision Records Sole architect of an architecture pattern and Rust reference implementation (8 crates, 155 tests) for governed multi-agent AI systems that treats every agent as Byzantine and keeps models outside the trusted computing base: Multi-Paxos control plane, Cedar and OPA policy engines, Model Context Protocol (MCP) server and client, Kafka transport, and a WebAssembly (Wasm) tool sandbox. Designed a policy admit path in which a formally verified Cedar floor can permit and a runtime OPA Rego bundle can only subtract, and a composite lease credential fenced at the Paxos log index so revocation and expiry need no quorum.

Designed information-flow labels on agent ingress with three floor-level prohibitions (read-then-send, steered destination, replayed content) that bound what a successful prompt injection can reach, independent of the model's own defenses.

Model-checked 16+ TLA+ modules under a mutation discipline in which each mutation must break exactly its property; discharged proofs in TLAPS and re-derived them in Lean 4, including a proof that the transcribed Rust admit decision is a step of the composition specification. Replayed all 82 recorded test traces through both the TLA+ and Lean 4 models with full agreement, and built a drift test that fails the build when any specification diverges from the code. Implemented sans-IO crash-fault Multi-Paxos with a write-ahead log, a Cedar engine pinned by digest, an OPA sidecar client with local Wasm fallback, and signed policy bundles. Authored the design corpus: 55 Architecture Decision Records, a Theory of Operation, an Adopter's Guide with conformance checklist, and a Reviewer's Guide. PROFESSIONAL EXPERIENCE

Distributed Systems Architect (Freelance) GLG (Gerson Lehrman Group) Oct 2026 - Present Independent subject-matter expert consulting on distributed systems, streaming data platforms, cloud infrastructure, and agentic AI architecture, with formal methods experience in TLA+ and Lean 4. Lead Software Engineer Capital One Sep 2014 - Oct 2025 Architected and deployed to production Capital One's first Docker platform: a data-analyst environment

(RStudio, Jupyter, Spark, Hadoop, HDFS, Kafka clients, SSO) running on Mesos, Marathon, and AWS EFS. Presented by Capital One at DockerCon; its successor remains in use. Delivered the first Capital One production Kafka ETL pipeline (Kafka, Kerberos, TLS, Apache Flume, Hadoop) moving data in Parquet, Avro, and JSON; later built Kafka Connect and Apache NiFi pipelines and migrated on-premises data to AWS.

Designed and implemented policy-as-code frameworks with Open Policy Agent (OPA) to enforce compliance and access controls across multi-tenant microservices, including fine-grained ABAC policies written in Rego and evaluated at runtime.

Led cross-functional engineering teams through full software development lifecycles in Python, Node.js, GraphQL, AWS Lambda, DynamoDB, and QuickSight, including a DynamoDB single-table design for a data- driven, responsive management UI.

Built real-time observability tooling with custom OpenTelemetry metrics, New Relic NerdGraph integration, and GraphQL schema stitching on a React, Node.js, Vega, and DynamoDB stack. Prototyped and evaluated LLM and RAG systems (SageMaker, ChromaDB, Milvus, Llama, Mixtral): vector- store selection, embedding and chunking evaluation, prompt design, and evaluation loops. Architected a prototype agentic solution (Ollama, Kafka, OPA, Paxos) that seeded TOPAK; performed price analysis of competing architectures to support build-versus-buy decisions. Software Engineer Acision (formerly Amteva, CMG, LogicaCMG) Nov 2000 - Sep 2014 Engineered core backend infrastructure and network automation for carrier-grade, highly available, geographically redundant telecom data transport; began as a Cisco Systems contractor and served as Solution Integration Engineer during the LogicaCMG years (2004 - 2007). Integrated and scaled Linux, OpenLDAP, and Cyrus IMAP for carrier-grade production message storage; automated bare-metal deployments on Sun Solaris and Linux via PXE and Kickstart. Managed production Sun ZFS storage through zero-downtime live migrations with no data loss; built automated test environments on VMware, Docker, CephFS, and ZFS. PATENTS AND PUBLICATIONS

Automated Streaming Data Platform - Co-Inventor, US Patent 11,025,688 B2 (Issued June 2021). Cloud-native streaming platform with automated provisioning, multi-availability-zone replication, disaster recovery, and zero- degradation live upgrades.

The Gripping Hand (grippinghand.dev) - Technical blog series introducing TOPAK and its components. ADDITIONAL EXPERIENCE

Bass Trombonist Richmond Symphony Sep 2000 - Present Professional orchestral musician. Performed with the Metropolitan Opera, New York Philharmonic, Minnesota Orchestra, Virginia Symphony, and North Carolina Symphony. EDUCATION

The Juilliard School - Music Performance The University of Georgia - Music Performance and Music Education WILLIAM SCOTT COCHRAN

Richmond, VA 804-***-**** *****.*******@*****.*** linkedin.com/in/scott-cochran-b069624 github.com/scottcochran November 16, 2026

Dear Software Engineering Hiring Team,

I am a senior software engineer and systems architect with 25 years of experience in carrier-grade and bank-grade distributed systems, event-driven architecture, and cloud-native infrastructure. Over the past year I built TOPAK (github.com/scottcochran/TOPAK), an open-source, formally verified architecture for governed agentic AI. I currently consult through GLG on distributed systems and agentic AI architecture, and I would like to bring that combination of production depth and rigor to your team. TOPAK combines TLA+ specification, policy as code in Cedar and Open Policy Agent (OPA) Rego, Multi- Paxos consensus, Model Context Protocol (MCP) connected agents, and Kafka transport in a reference implementation of eight Rust crates and 155 tests. Sixteen TLA+ modules are model-checked, the proofs are re-derived in Lean 4, and every test trace replays through both models. Before TOPAK, at Capital One I prototyped and tested large language model (LLM) and retrieval-augmented generation (RAG) solutions using SageMaker, Python, ChromaDB, and Milvus with Llama and Mixtral models, including vector database evaluation, embedding tuning, chunking evaluation, system prompt design, and evaluation loop design. I have developed MCP servers from scratch and integrated agents with them using cloud-based models, and I am an enthusiastic user of Claude Code.

At Capital One I was the first to introduce Kafka and Docker, integrating both with the Hadoop cluster. I developed an analyst sandbox hosted on Docker, Mesos, and Marathon consisting of RStudio, Apache Spark, Jupyter Notebook, HDFS, NFS, Kafka and Hadoop clients, and single sign-on. Capital One presented it at DockerCon, and its successor is still in use. I developed and deployed extract, transform, load (ETL) solutions for on-premises Hadoop ingestion using Kafka, Kafka Connect, Apache NiFi, and Apache Flume, and migrated data from on-premises to AWS and, within AWS, from Snowflake to PostgreSQL and DynamoDB using Python Lambdas built on Pandas and NumPy. I have also designed and developed custom metrics solutions end to end: database selection, high availability, geo-redundancy, schema design, ETL, and visualization. That work included OpenTelemetry metrics and New Relic NerdGraph integration with GraphQL schema stitching in a React, Node.js, Vega, OPA, and DynamoDB stack for real-time monitoring and alerting. Its DynamoDB single-table design enforced fine-grained attribute-based access control (ABAC) policies written in Rego and evaluated at runtime against state. I am experienced in REST API development and GraphQL schema design, and in statistics and visualization tooling including scikit-learn, NumPy, Pandas, Vega, Grafana, and QuickSight. Earlier, in carrier-grade telecom at Acision, I developed bare-metal deployment solutions for Solaris and Linux, administered NFS, HDFS, Lustre, ZFS, EFS, and CephFS through live hardware, network topology, and interconnect upgrades and migrations, and integrated and supported PostgreSQL, DynamoDB, Accumulo, OpenLDAP, and Cyrus IMAP.

I would welcome a conversation about how this experience maps to your systems. Sincerely,

William Scott Cochran



Contact this candidate