Dhavan Shah
Fail-closed budgets, approval gates, evaluations, and grounded company memory — installed before agents run unsupervised.
UZH MSc · Kaggle Top 0.5% · Independent practice · Building KazenAI
Fixed-scope work for Series A to C teams who already feel agent cost or trust risk. You get written deliverables, not open-ended billing.
$3,500–$7,500 · 1–2 weeks
Architecture and workflow review. Cost and loop risk. Tool permissions. Eval and failure-mode report. Budget and approval recommendations. Prioritised 30-day roadmap and a walkthrough.
$10,000–$25,000 · 4–8 weeks
One controlled production workflow: pre-call budget enforcement, permission policy, human gates, claim checks, audit trail, evaluation, and deployment notes.
from $3,000/mo · typically $3–8k
Architecture reviews, evaluation maintenance, cost incident analysis, policy updates, and limited implementation when new agent workflows appear.
Multi-workflow integrations, private deployments, and three-to-four-month reliability programmes are scoped separately. Typical engagements begin around $25,000 and may exceed $75,000 depending on system complexity, responsibility, and production requirements.
Fit check
Scoped work is strongest when there is a production or near-production workflow, a technical owner, and enough signal to audit behavior.
Approach
A clear sequence so you know what you are buying before any code lands in your stack.
Thirty minutes on production risk, stack, and whether an audit or pilot is the right first step.
No obligationWritten findings on spend loops, gates, evals, and memory. The product is the report, not a slide deck.
1–2 weeksOptional. Harden one workflow with fail-closed budgets and evidence. KazenAI patterns when they fit your stack.
Scoped deliveryOnly when the same need repeats: ongoing reliability work, or design-partner modules (FinOps, Brain).
OptionalThree reference systems behind the practice. Engagements install the same patterns in your stack.
Why teams trust this work
I am an independent AI Systems Engineer and founder of KazenAI, focused on making production agent workflows controllable: budgets, permissions, approval gates, grounded retrieval, evaluations, and replay.
Previously at DevRev, I shipped LLM systems across live customer workflows. My broader experience includes applied AI work with AXA XL and Sulzer in Zurich, an MSc in Data Science from the University of Zurich, and Kaggle Competition Expert, ranked in the global top 0.5%.
I work through fixed-scope audits, production implementations, retainers, and selected fixed-term AI systems contracts. KazenAI remains my background technology and reusable IP; client code and confidential information remain segregated under the engagement agreement.
I take a small number of engagements. Share a short brief and I will reply within one business day. If there is a fit, I will send a private link to schedule a 30-minute call.
Prefer email? [email protected]
