Dory
From United States (UTC+2)
Lemon.io stats
3
offers now 🔥Dory – Kubernetes, Python, AWS
Dory is a Staff-level AI and infrastructure engineer with 20+ years of shipping production systems across the US, Europe, Israel and Australia, from Microsoft R&D to co-founding engineering on an LLM routing platform. He takes vague goals from founders and clients and owns them end to end, from architecture to Kubernetes in production. His core stack is Python, TypeScript/Node.js, Kubernetes, Helm, Terraform/Pulumi, AWS and GCP, with deep hands-on work in LLM routing, voice AI and agent orchestration. He helped grow the valuation from $70M to $250M+, and at his current AI startup he built a sandboxed agent workflow where engineers run 10–20 AI agents in parallel.
23 years of commercial experience in
Main technologies
Additional skills
Direct hire
PossibleReady to get matched with vetted developers fast?
Let’s get started today!Experience Highlights
Senior Staff AI Engineer
An AI coaching platform for enterprise contact centers that trains, assists and evaluates customer-support agents. Agents practise on realistic AI simulations of real customer calls, get real-time guidance during live calls, and receive automated QA and coaching afterwards. Because it processes live voice at scale, speech-processing cost, latency and reliability directly shape the product.
- Built a Slack-driven agentic development workflow: each request spins up an isolated cloud VM sandbox where an AI agent builds the feature and opens a pull request, and the sandbox is torn down on merge. Engineers run 10–20 agents in parallel, and the CEO, product managers and designers ship reviewed changes to production every week;
- Put guardrails around the agents: restricted tool sets, tool-call limits, prompt pre-filtering, PII cleaning, automated checks and human review before merge;
- Cut voice-processing costs 5x by moving speech-to-text and text-to-speech from a hosted vendor to a self-hosted Deepgram cluster on Kubernetes, and built a Helm volume-rotation approach that adds new language models without reinstalling or taking the cluster down;
- Improved platform performance 10–15x and redesigned the customer sandboxes and their infrastructure, letting the sales team build customer simulations that helped unlock millions in revenue;
- Replaced a brittle, manual deployment pipeline with Kubernetes-based infrastructure on GCP, enabling zero-downtime releases and lowering infrastructure costs;
- Kept voice workloads running through GPU shortages with GPU autoscaling, fallback deployments, shared GPU capacity and pre-provisioned resources;
- Hired a three-person engineering team and designed the AI engineering hiring process, including screening questions and practical exercises.
Staff Engineer (Co-Founding Engineer)
An LLM routing platform that sits in front of many language models and sends each prompt to the best-fit model, balancing cost, output quality and latency. Enterprises use it as a single API across providers such as OpenAI and Anthropic, including in on-premise deployments, so it must handle high-throughput, low-latency traffic and keep working when a model provider goes down.
- Helped grow the company's valuation from $70M to $250M+ and close multi-million-dollar enterprise deals by owning the core platform and the proof-of-concept work with customers, working directly with the founders;
- Redesigned and built the core LLM router that takes OpenAI and Anthropic traffic and picks the best model for each request on cost, quality and latency, working at the protocol level (streaming and chunking strategies), with provider failover and pre-hoc and post-hoc routing;
- Architected the high-load microservices and the enterprise on-premise deployments (Helm charts for air-gapped environments), and migrated all systems to Kubernetes, improving throughput and latency by 500%;
- Built evaluation pipelines that benchmark routing decisions across providers and measure output quality, working daily with ML researchers, plus Kubeflow-based tracking of model performance and routing accuracy;
- Built a high-volume ClickHouse store that records every inference request and response for fast retrieval and analysis;
- Load-tested model inference under production traffic patterns to find breaking points before customers did, and redesigned and open-sourced the async model adapters library that calls many LLM providers through one interface;
- Built a 10-person team across infrastructure, frontend, backend and product, including the hiring process for senior engineers.
Staff Engineer / Tech Lead
An enterprise carbon-management platform that helps large companies measure their emissions across operations and supply chains and plan reduction actions. Built inside a global consulting firm and then spun out as a standalone product, it pulls large volumes of data from many sources for each customer, so onboarding speed and data reliability decide how quickly new enterprise clients can start using it.
- Cut customer onboarding from hours to minutes by bringing together a team of new hires to rebuild and streamline the ETL pipelines;
- Led the platform's move from MVP to its second major version during the spinoff, owning the technical work from architecture through delivery;
- Built serverless data processing on AWS Lambda that stays cost-efficient at scale;
- Found and fixed several business-impacting data issues, cleaning and standardizing incoming data and handling source-specific processing rules;
- Scaled the supporting systems, from CI/CD pipelines to database sizing.
Lead Engineer
A direct-to-consumer health and wellness brand for older adults in the US, selling personal care products such as bladder protection and skincare online and through retail partners. Its customer-facing web apps and customer-support systems have to keep running smoothly as order volumes grow.
- Took the lead on a three-month migration of customer support from Zendesk to PureCloud within two weeks of joining, designing the architecture and data model across three Python/Django and React apps and delivering before the vendor cutoff;
- Saved $60K a year by moving feature flags to an open-source platform, and trained all product managers and engineers on it within a week;
- Improved application performance 15x and refactored the apps with zero downtime;
- Mentored two junior developers.
Senior Software Engineer
The advertising systems behind a major US media group's roughly $100M advertising business, used to sell, manage and deliver ad campaigns. The systems carry the revenue directly, so any instability during the busiest sales season hits income.
- Took sole ownership of every software product after the previous development team left, stabilizing the systems to protect revenue through the busiest season;
- Rebuilt the advertising systems on AWS Kubernetes, splitting the monoliths into 5–6 REST microservices and merging overlapping libraries;
- Migrated the frontend from Angular.js to React;
- Sourced and hired the international development team that took over the systems.
Senior Software Engineer / Team Lead
A custom ERP for a US software development consultancy with offices in New York and San Francisco. It automates billing, hours tracking and resource planning across the firm's engineering staff, replacing manual back-office processes.
- Built the ERP together with the CEO, taking it from an MVP in a few weeks to a product used daily by 160 engineers;
- Automated the consultancy's billing, hours tracking and resource planning on Node.js, Ember.js and MongoDB;
- Hired and led a team of 10 across design, UX, development, project management, DevOps and testing;
- Ran live coding sessions for engineers in the New York and San Francisco offices.