Cyrus Nouroozi

Superhuman scientific evidence synthesis

DSPy is the project for programming — not prompting — LLMs, and is Stanford NLP’s #1 project on GitHub. Core contributor towards DSPy 3.0 — 1. Architect of Adapters. Adapters are the components that map signatures to prompts prior to optimization, and parse responses into data structures. 2. Led/implemented V1 of async support with of dspy.syncify and dspy.asyncify 3. Led/implemented V1 of streaming support with dspy.streamify 4. Optimizer research

Helping push the frontier of AI or Web3 on 0–1 projects. Software Development 1. Open-source django backend template sponsored by LlamaIndex Smart Contract Development: 1. Multi-chain deployments @ syndicate.io, velodrome.finance 2. Modular policy engine @ predicate.io Strategic Advisory: 0. Technical GTM for DeFi compliance startup @ 0xbow.io 1. GTM for carbon dioxide removal crypto-carbon credits @ electrictree.com 2. Inventing regenerative stores of value @ spirals.so

Invited to join the leading multi-agent AI research lab camel-ai.org at a PhD level. 1. Reading papers on existing state-of-the-art methods in large-scale language models and multi-agent systems. 2. Brainstorming and proposing new ideas or modifications to existing methods 3. Implementation, Experimentation, Evaluation, Documentation, and Reporting

Rotational program towards a full-time offer as a senior engineer on a combined IC and Managing workload, left to begin my traveling artist phase at the intersection of crypto and climate change. 1. Led the 0–1 development and devops of the backend of Syndicate Protocol with 2 full stack developers from conversation from scoping to delivery 2. Co-authored from scoping to production the smart contracts for Syndicate Protocol with Conner Swenberg 3. Managed a team of 4 frontend developers in Kenya and finished our sprint early. 4. Ideated, championed, and developed the Polygon smart contract deployment of Syndicate Protocol. Most Syndicate Clubs are now launched on Polygon due to its lower costs.

Pro bono. AI mental health chatbot available to 19MM+ people with studies finding a reduction in depression symptoms of 28% and anxiety symptoms of 18%. Ideated, championed, and prototyped a feature to delight users: A Python module using GPT-3 to safely summarize and reflect the complex multi-sentence emotions of people opening up about how they're feeling. Abstract: X2 develops Tess, a therapeutic chat bot that is capable of handling and reducing symptoms of mild cases of anxiety and depression by coaching people through reflective exercises. Kash Money Records Ltd. was brought on to explore the use of GPT-3 in enhancing the user experience. Intentionally induced creative mind states resulted in the insight of using GPT-3 to perform active listening on a user’s description of their current emotional state. Active listening was chosen as it is an accepted therapeutic practice that makes patients feel heard and understood. Through process mapping, user journey mapping, role playing, prototyping, and iterative design, the Engineering Design Process was used to develop and validate low and medium fidelity prototypes. A medium fidelity prototype using GPT-3 proved successful at accurately generating sentences that mirrored complex patient emotions. Internal testing of active listening delighted the senior leadership team and earned support to A/B test the impact with live user testing.

Wombo was downloaded 100M times in its first 6 months, and has ranked #1 on the App Store in 20+ countries. + Napkin-sketcher + Product hustler + UX designer + Analytics ninja + Growth guru + Founding CTO + DevOps / Backend / Mobile Developer

I took Conversify from a Jupyter notebook running on the founder's computer every month to a serverless cloud browser automation for sales outreach, and we scaled from 20 users to nearly 200. Learned a bunch about AWS: Don't use it until you need to, it only slows you down. Browser automation: You can scrape nearly anything, especially if you hack into the hidden developer API behind frontend application frameworks like React and EmberJS. Conversify was admitted to NextAI's 2020 batch, Toronto's best accelerator for AI companies.

During my time at TimeSaved I crafted a developer-centric React component library using styled-components. Provided with Figma designs, I delivered a complete design system in 7 weeks. This library was so user-friendly that our Head of Product felt empowered to update the UI himself. In the interest of development velocity, sometimes I would wrap other React libraries. Sometimes I would write my own. Once I had to write a build script to transform the SVGs of our icon library into JSX. In the end, we had on-brand components for: Text Fields / Autosizing Textarea Enhanced Text Fields (e.g. with Google Maps autocomplete, or a datepicker on click), Checkboxes / Radio Buttons / Switches / Sliders Buttons / Dropdowns Typography / Icons Layout / Tables Later, I was tasked with resuscitating a project that had been developed by a previous contractor. It was a React Native app that was "70% done". The app wouldn't compile, so I switched us over to Expo, learned some React Native / Redux, integrated it with a GraphQL server, refactored a bunch of old code, built out new screens, and published it to the app stores.

Docker, ETL pipelines with Python / Pandas, and productionizing data science that resulted in millions of annual revenue for Charles Schwab.

Most impactful developer on the team as measured by story points completed per month. First responder for bugs; responsible for 90%+ of fixes. Crafted end-to-end features with SCSS, Bootstrap, Rails. Improved page load time through a variety of methods. I was DevOps. Recognized for excellent code quality. Anticipated the need to manage collaborations across social networks and spearheaded the V2 of our collaboration management software. As project owner I consulted stakeholders and customers to take V2 from 0 to 1 using Sketch, jQuery, SCSS, Rails, PostgreSQL, and TDD. Increased active campaigns per account manager — a company-wide KPI — by 40%. Onboarded an all-new 4-person engineering team. Noticed a breakdown in flow during design-developer handoff, performed root cause analysis, and established developer consultations as part of the design process. Eliminated problems rooted in a lack of technical context. Built a multi-social network scraper from scratch in Elixir to power ad performance analytics, which provided the data for an analytics dashboard (previously: a manual wrap-up report) . As project owner I consulted stakeholders, architected, and developed using TDD. Backed by MongoDB. Deployed to AWS using Docker and Jenkins, integrated with DataDog, Scaled from 1,800 to 18,000 requests per hour. Recognized the opportunity for a runoff effect and added 50,000+ creators to our onboarding pipeline (previously, 150+). Partnered with a teammate from the Creator Acquisition team to develop a Virtual Assistant pipeline to process scraped social network data.