We architect and ship production-grade systems across blockchain, AI/ML, and distributed systems — for clients where precision is not optional.
Tripod operates at the hardest intersections of the modern stack — where academic rigor and production engineering are both required to get it right.
End-to-end pipelines, analytics platforms, and cross-domain data systems built for scale, compliance, and correctness. From ingestion architecture to federated query layers that span Web2 and Web3.
Protocol-level development through production trading systems. Multi-chain architecture, KYC/AML compliance infrastructure, smart contracts, and institutional-grade digital asset platforms built to survive regulatory scrutiny.
Production ML pipelines, multi-agent systems, and multi-modal models — designed with the rigor of academic research and the discipline of production engineering. We evaluate systems the way an adversary would.
Fault-tolerant, real-time systems engineered on Erlang/OTP principles. High-concurrency backends, event-driven architectures, and platforms where uptime and consistency are non-negotiable constraints.
A cross-section of engagements: product builds, client platforms, academic research, and internal capability development.
Companies operating across blockchain and traditional systems spend 40–60% of their data engineering time on plumbing: one-off chain indexers, warehouse syncs, and manual joins. 3pods is a single platform that ingests, normalizes, and serves blockchain events, relational data, and API records as one queryable surface. Analysts write a single SQL statement that spans both worlds. Compliance teams get cross-domain alerts that fire when an off-chain signal correlates with on-chain activity.
Deal sourcing and investor qualification for a PE firm
Eliminates manual sourcing across brokerage listings, federal datasets (IRS 990s, CMS NPPES, BLS QCEW), and state registries. A two-track ingestion platform produces a deduplicated, scored, queryable entity graph of opportunities — with a manifest-first plugin model that scales to arbitrary sources without modifying core infrastructure.
National-scale cardiac arrest analysis with UNLV researchers
ML pipeline identifying distinct comorbidity phenotypes across 47,500 in-hospital cardiac arrest patients from the Nationwide Inpatient Sample. Adaptive PCA/NMF, K-Means and GMM clustering, FDR-corrected mortality testing, and mediation analysis reveal a measurable shift from cardiac-dominant to infection-dominant patient profiles during the COVID era.
RCT platform measuring trust recovery in failing AI advisors
A 2×2 randomized controlled trial examining whether users recover trust in an AI advisor after visible accuracy failures — and which design choices mediate recovery. Custom Phoenix LiveView platform captures sub-millisecond trial responses, trust trajectories, and behavioral compliance across 40 trials per participant. IRB-aligned, JARS-Quant reporting.
Capability research into LLM detectability in evaluative knowledge work
A multi-stage agent pipeline that produces evaluative text designed to survive real-world AI detection heuristics — framed defensively. The adversarial scrub loop exposes weaknesses in existing detectors and points the way to stronger authenticity infrastructure. Dual-use capability research with disciplined methodology and 120+ tests.
A targeted attack arrested at the last moment — then reverse-engineered completely
A fabricated executive persona made credible contact through a professional network under the cover of a consulting opportunity. The payload arrived as private repositories containing a novel attack pattern: editor workspace poisoning — repository-tracked editor configurations that silently relocate the git hook directory to execute a remote payload on project open, with no prompts and no visible output. Attack arrested before code execution. Full investigation produced 172 evidence artifacts with SHA-256 integrity manifests, attribution of approximately 40 sibling lure repositories to a single operator identity with timezone reconstruction and commit-burst analysis, and coordinated abuse reports filed with package registries, source-control platforms, and federal cybercrime authorities. A complete, reproducible case study for supply-chain threat modeling and defender training.
Tripod doesn't just build production systems — it evaluates them the way a researcher would. That discipline shows up in every engagement.
Doctoral research in progress at the University of Nevada, Las Vegas — sitting at the intersection of Tripod's clinical ML practice and cutting-edge AI research. An ongoing source of methodology that feeds directly into engagements in healthcare and AI systems.
Both active research collaborations (IHCA phenotyping, AI trust RCT) are designed to STROBE and JARS-Quant reporting standards — peer-review-ready. Clients in regulated or high-stakes domains get analysis they can defend to a scientific audience, not just an executive one.
Work spanning DoD contracting (principal holds prior TS/SCI CI polygraph clearance from government contract work), healthcare enterprise software (HCA), SEC-compliant securities infrastructure (Realio), and IRB-governed human-subjects research. Tripod understands what operating under regulatory scrutiny actually requires — not as a compliance checkbox, but as a working constraint.
The Turtest project and the supply-chain forensics investigation reflect the same discipline: testing systems the way an adversary would. That adversarial orientation — applied to LLM detectability, developer supply chains, and trust dynamics — is what clients in high-stakes domains increasingly need from technical advisors.
Graham Haley founded Tripod Consulting in 2015, bringing over fifteen years of accumulated engineering experience to bear on a simple operating principle: apply the same rigor to a client's hardest problems that you'd apply to your own. That principle runs every engagement.
The range — blockchain at the protocol level, production ML pipelines, fault-tolerant distributed systems in Elixir/OTP, academic-grade statistical analysis — isn't breadth for its own sake. It reflects the reality that the hardest technical problems don't respect domain boundaries. The IHCA study required ML methodology and healthcare data engineering in the same breath. The DealFlow platform required graph data modeling and domain-specific scoring. 3pods requires distributed systems, blockchain integration, and ML — under a single architecture.
Tripod operates at the intersections where those disciplines meet. That's where the work gets interesting.
If you're working on something technically ambitious — in data infrastructure, blockchain, AI/ML, or distributed systems — let's discuss whether Tripod is the right partner.
info@tripod.consulting →