I build reliable, production AI systems—fast.

CTO at Techjays. I turn messy, high-stakes workflows into dependable AI products with clear SLOs, human-in-the-loop guardrails, and measurable ROI.

See selected work Browse resources

99.95% Availability SLO

p95 ≤ 1.2 s Realtime voice

11 / 11 Evals passing before release

What I do

Explore my approach

I partner with CXOs and functional leaders at enterprises to ship applied-AI products that survive real traffic and scrutiny—then I harden them for scale.

Build

Agentic workflows, “talk-to-data” assistants, retrieval-augmented reasoning, and realtime/voice experiences.

Make it dependable

SLOs, multi-model routing, circuit breakers, tracing, and cost budgets baked in from day one.

Keep it safe

Human-in-the-loop decisions, audit trails, privacy posture, bias checks, and release gates.

Products we run at Techjays

HireFinch — AI voice interviewer

Structured, rubric-based interviews with cited transcripts, agentic “talk-to-data” cohort insights, and multi-model failover.

Outcomes: fewer low-signal screens, faster panel prep, and transparent scoring that references the exact transcript spans behind each rating.

Highlights: provider-agnostic inference shim, circuit breakers and health checks, webcam-assisted proctoring with short-retention images, and SLOs for latency, ASR accuracy, and availability.

See how HireFinch works

SparkFinch — a humane alternative to timesheets

Voice-to-voice team journaling that turns daily updates into manager-ready rollups—without adding bureaucracy.

Highlights: organic adoption (no mandate), manager usage as the KPI, and privacy-aware defaults. SparkFinch helped cement Techjays’ shift to an AI-first org alongside Claude-centric project memory and the “SEAL-6” pilot program.

Read the operating notes

Selected work (anonymized)

Read the build notes

Enterprise LLM Copilot for Ops, Freight & Finance

Agentic workflow + RAG over OMS/TMS, email, and knowledge bases; safe tool schemas; PII scrubbing; audit logs; OTel traces.

−45% scale-ticket chase emails.
Invoice cycle time −1.3 days (~−28%).
Exceptions 12% → 6% (−50%).
~10.4k hours/year saved (~$520k/year at $50/hr).
SLOs: p95 1.2–2.0s, 99.95% availability; ~$0.011/action.
Live task success 91% with human oversight.

Read-only posture, HITL gates, weekly red-team, canary releases, and rollback round out the governance.

Read the build notes

AI-Powered Freight Quoting Co-Pilot for Logistics

End-to-end quoting assistant for a national logistics provider—classifies RFQs, extracts shipment details, retrieves policy, reasons about pricing, and drafts explainable quotes under HITL guardrails.

Why it works: retrieval over memorization, deterministic pricing calculators, node-level evals with adaptive routing, and full auditability.

Read the Techjays case study

Want more work samples?

I keep a larger gallery of anonymized case studies over on the Techjays site—go deeper on evaluators, red-teaming, and governance patterns there.

See more Techjays case studies

Past work before Techjays

Akamai — internal data monitoring at scale

Worked on telemetry and alerting flows for massive infrastructure: high-cardinality data, noisy signals, near-real-time extraction.

Built empathy for designing reliable systems that stay observable when volume spikes.

Google — developers.google.com

Contributed to the developer site experience, sharpening a bias for clear information architecture and accessible documentation.

That lens still guides how I communicate complex AI systems and tooling.

Browse more work

GSS Pinnacle Award — Akamai

One of five global managers recognized for leadership that keeps Global Services & Support seamless.

The Pinnacle Award is the highest honor in GSS, celebrating leaders who deliver excellent results, mentor deeply, and empower teams to keep customers first. It reflects how I build high-trust teams and scale execution under pressure.

Assemble Effective Teams Bottom-Line Results Customer First Empowering Mentorship

View verified badge

MIT Sloan — Mastering Design Thinking

Executive program focused on framing ambiguous problems, prototyping rapidly, and scaling solutions with measurable outcomes.

This course sharpened how I translate messy stakeholder input into experiments grounded in evidence. It complements my GenAI build process with structured discovery, fast customer validation, and disciplined iteration loops.

Customer insight Experimentation Prototype rigor

View credential

How I build dependable AI

Discovery → Pilot → GA (with gates)

I start in the user’s workflow, ship a HITL pilot, then add release gates for accuracy, latency, privacy, and fairness before GA. A small “SEAL-6” team stress-tests tools; we prefer organic pull over mandates. Claude-centric project memory keeps PM, design, and engineering in sync; voice capture removes friction where typing is a chore.

Reliability & cost

Multi-model routing with circuit breakers, schema-normalized failover, per-step tracing and cost metering, strategic caching, and explicit SLOs/SLIs keep systems predictable.

Safety & governance

Humans own the important decisions. PII minimization and redaction, audit logs for prompts/tools/versions, parity checks across cohorts, periodic red-teaming, and clear rollback playbooks are standard.

Evals

Offline sets plus live disagreement tracking. Gates for rubric adherence, evidence precision/recall, calibration, fairness deltas, and proctoring performance. HireFinch and the Ops Copilot ship with automated CI gates and canaries.

Connect elsewhere

Prefer to follow along on social channels? These feeds focus on reliability patterns, eval tooling, and applied-AI operator stories.

Writing & resources

AI-Powered Freight Quoting Co-Pilot

A practical blueprint for turning RFQ chaos into analyst-approved quotes with explainability and cost control. Includes sequence diagrams, eval suites, and a rollout plan from HITL to progressive autonomy.

Get the resources

Techjays’ AI-first transformation playbook

What worked—and what didn’t—when evolving Techjays: Claude Projects as shared memory, SEAL-6 pilots, and voice journaling that leaders actually use.

Highlights and excerpts stay in private briefings; there’s no public download.