Amir Behbehani Email & Phone Number

San Francisco, Ca, Us

As the founder of Memra, I spent the first year and a half developing Large Action Models (LAMs), the core of our AI-driven platform. These digital employees go beyond task automation by managing entire jobs, dynamically breaking down complex tasks into subtasks and interfacing seamlessly with diverse systems and APIs.Collaborating with an engineering partner, I architected and developed Memra’s core AI system, ensuring it is scalable, robust, and adaptive. My focus has since shifted to market deployment, overseeing integration, strategic partnerships, and advancing our market positioning. I also advocate for Memra’s solutions through podcasts and papers, engaging with the AI community to drive thought leadership.🔹 Key Contributions:• Built LAMs to autonomously manage entire workflows and system interactions.🔹AI System Architecture: • Built a proprietary vector DB to store knowledge graphs in vector space, modeling context beyond RAG.• Created an info-theoretic model for task reasoning and decision-making.🔹 Strategic Growth & Market Positioning:• Lead efforts to integrate our digital employees into client operations, focusing on strategic partnerships and market expansion.🔹 Thought Leadership:• Promoting Memra’s solutions through podcasts and industry papers.As a founder, I combine deep AI expertise with strategic initiatives to bring Memra’s groundbreaking digital employee technology to market.

Serial Metrics, founded in 2010, specialized in developing core AI intellectual property. The company partnered with early-stage startups. At Serial Metrics, I played a key role in both the technical development and leadership of the company. I led the growth of our team, fostering a culture of innovation and technical excellence. I was actively involved in developing the AI solutions that formed the core IP, directly contributing to the technology that drove the success of our portfolio companies. My work focused on building scalable AI systems that addressed real-world challenges for our partners.These efforts positioned Serial Metrics as a valuable partner to startups, resulting in 15 major acquisitions by companies such as Google and Meta.Portfolio:• Levo• Leaf Logistics• Orion• Medibio • CloudLending (Acquired by Q2)• Metadata• Zuli (Acquired by Google)• Waypoint (Acquired by Invitation Homes)• Jobr (Acquired by Monster)• 8Tracks• Lynda (Acquired by Linkedin)• Identified (Acquired by Workday)• Atypon (Acquired by Wiley)• Path (Acquired by Adecco)• Zipongo (Now Foodsmart)• Swaylo (Acquired by Meta)• Eventbrite

San Francisco Bay Area

In an advisory role with Levo, I achieved the following key milestones:• LevoSearch Creation: Spearheaded the development of a natural language query engine to enhance user experience and streamline data access for healthcare professionals.• Technical Leadership: Led a team and coded a C++ data pipeline, achieving seamless integration with enterprise databases and ERP systems and enabling real-time analytics.• NLP Deployment: Utilized natural language processing to convert user queries to SQL in real-time, delivering actionable insights into employee performance and engagement with 95% accuracy.

Helped define a forward derivatives model that doubled freight market liquidity and reduced deadweight- loss by 50%, optimizing pricing and network efficiencies for the freight logistics industry.• Employed machine learning techniques (SVMs, Random Forests, and GBT) to analyze transportationdata and forecasted supply and demand variables with 90% accuracy, yielding a 20% efficiency gain.• Helped grow company value by 150% from $100MM to $250MM after implementing a forward derivativepricing structure, improving revenue stability for thousands of shippers and carriers.• Achieved 10x YoY net revenue growth through an economic study using Structured Equation Modeling.

San Francisco Bay Area

Launched an auto-ETL platform and directed the company to rapid growth, successful enterprise sales, and a 1000x return on initial seed investment.• Created an NLP-based ETL system for real-time data analysis, reducing query creation times by 90%.• Devised a neural network-based language model that transformed databases into knowledge graphs, enabling real-time analysis and NLP queries, reducing analysis time by 50%.• Played a key technical role in enterprise sales efforts, contributing to a 300% YOY increase in revenue.• Increased adoption by 25% and satisfaction by 40% via cross-functional collaboration to improve the platform.

Pioneered an AI platform using abnormal, arrhythmic cardiac signals collected from wearable devices to aid in clinical diagnosis, resulting in an average of 8 years earlier diagnosis.• Boosted Alzheimer's diagnosis accuracy by 73% with an AI platform that analyzed cardiac signals fromwearables, reducing inter-diagnostic variance from 55% to over 95% using random forest models.• Coded a C++ application to convert EDF binary data to CSV format, enabling faster and more accuratedata preparation and analysis, improving model development speed by 2x.• Published ML results in a peer-reviewed neurology journal, contributing to $17M in venture financing.

Spearheaded the launch of a transformative AI-powered lending service, reducing the loan process from months to minutes, resulting in the company's acquisition value of $105 million.• Predicted default likelihood using non-parametric models, improving risk management by 30%.• Integrated ML models to Salesforce using REST API and Apex, increasing market reach by 100%.• Deployed ETL flows for financial platform interactions, reducing customer onboarding time by 50%.

San Francisco Bay Area

Developed an AI-assisted targeting system that resulted in a 10x return on ad spend and generated a new pipeline of $3 million in potential revenue.· Designed and developed a data-driven system that merges CRM sales data with Google ad campaigns to identify effective campaign attributes for driving sales· Implemented real-time data integration and analysis to identify the attributes of ad campaigns that drive sales, including target audience, ad copy, images, and call-to-action· Built a predictive model using machine learning algorithms to determine the likelihood of a campaign driving sales based on CRM data, with the ability to update the model in real-time for continuous optimization· Integrated the system with Salesforce CRM through Python libraries such as simple-salesforce, resulting in efficient prospect targeting and improved campaign results, and maximum conversion rates.

Launched an initial product offering that hastened the path to securing a 2.5 million dollar round of funding.· Developed and implemented a cross-label music recommendation engine utilizing machine learning algorithms and data analysis, resulting in improved user music discovery and increased engagement across different music labels.· Enhanced the system to take data from concert attendees, which allowed for cross-selling into new audiences and increased revenue for the music labels.· Platform developed using C++, PHP, Python, and JS stack with a Postgres backend.· Designed system using a microservices architecture and deployed on cloud infrastructure for scalability and reliability.

Discovered, patented, and sold a successful home automation system to Google, resulting in a 40x increase in company valuation from $500,000 to $20 million.• Designed 99% accurate presence detection system for adjusting home appliances by user location.• Prototyped machine learning code in Python and ported code to Objective C for the Zuli iPhone app.• Optimized algorithm for 2x power and memory efficiency while streaming gyroscopic data.

Developed an automated valuation system using deep learning models on publicly available data, including the number of bedrooms, sqft, county, yard size, year built, etc., to estimate property values for portfolio acquisition. This eliminated the need for “walkthrough” property inspections, where – for example – a property inspector evaluates worn kitchen cabinets or missing roof tiles. The process is time-consuming and, itself, costly. More importantly, physically, inspecting properties is an unscalable process for a company whose business model requires time-sensitive, bulk home purchases.Waypoint Real Estate Group LLC, buys, fixes, and rents foreclosed homes, nationwide, and is among the largest investors in the U.S. real estate market, currently owning upwards of 10,000 properties. In February 2014, Starwood Capital Group acquired Waypoint and spun out as Starwood Waypoint Residential Trust (NYSE: SWAY).

San Francisco Bay Area

Developed a unique matching algorithm, leading to a quick sale to Monster.com with a 10x valuation increase and a $12.5 million dollar exit.· Developed a matching algorithm to optimally surface “Tinder-like” matches between job seekers and recruiters.· Created a random forest model to match job seekers and job posters by presenting candidates and recruiters with members of the other population that (1) they are most likely to be interested in and (2) are most likely to reciprocate that interest.· Utilized machine learning techniques to analyze large amounts of data from job seekers and recruiters to identify the attributes and conditions that lead to successful matches.· Integrated the algorithm with the company’s job platform, to create a seamless and efficient experience for both job seekers and recruiters.· Achieved a match rate of 90% by identifying the most suitable job and candidate matches, improving the overall job placement process.

San Francisco Bay Area

Developed core music recommendation engine, driving 8tracks from start-up to $50M valuation.· Engineered a recommendation system using R composed of three different models and modeling approaches resulting in a system that recommends playlists to users.· Category 1 Model: Utilized the dplyr and tidyr libraries to pull and organize user data by location, and the recommenderlab library to run a tag-recommendation algorithm on the data to identify the most popular tags within a given user’s region.· Category 2 Model: Utilized the caret and arules library to create a set of 20 ‘top recommendations’ by user with their associated lift statistics, by analyzing the ‘cross-section’ of mixes that users in a given demographic/geographic position “Like”.· Category 3 Model: Utilized the randomForest library to take a different approach by using behavioral data, and creating a recommendation score that can be used to generate customized lists of tags for a user. Additionally, utilized the dplyr and tidyr library to work with data, and caret library to train and test the model.

Developed machine learning video recommendation system with clustering, enhancing IP and aiding $1.5 Billion dollar sale to LinkedIn. · Implemented a video recommendation system using machine learning to analyze usage data from both active and canceled users· Utilized clustering algorithms such as K-means to group users based on their viewing patterns and identify user segments· Applied collaborative filtering techniques such as matrix factorization to predict which videos users in each segment are likely to watch next· Improved the accuracy of video recommendations for users by personalizing them based on their segment· Provided valuable insights for the marketing department by identifying patterns that emerge across sets of users, such as which videos are most popular among users who have canceled the service, and which videos are most likely to retain users.· Preprocessed and cleaned data, transformed and normalized it to ensure it is in a format that can be used for analysis· Utilized python libraries such as Pandas, scikit-learn and libraries for natural language processing for data manipulation and analysis.

Developed the Geo-Inferencing system for Identified.com for user location detection, boosting IP and contributing to sale of the company to Workday.· Created a geo-inferencing system to detect where users live, down to the metropolitan statistical area. The technology helped spur the sale of the company to Workday in 2014.· Developed a geo-inferencing system, able to infer the location of users that had not explicitly stated their location, using Facebook user data ingested from Hadoop and leveraging a set of random forest models trained on those data to predict user locations within a metropolitan statistical area. · Generated complex SQL queries to convert relational data sets into a dyadic data format (graph data set) to ensure the statistical model incorporated interaction effects instead of the more traditional statistical methods of controlling interaction effects. · Leveraged Census Bureau data sets to cluster Lat/Long data according to a defined set of Metropolitan Statistical Areas to create a predictive set of user locations. · Once deployed, the production system could take 700,000 Facebook accounts with known location data and infer the location of 50 million accounts with no corresponding location data. The deployed models performed at approximately 95% accuracy.

Developed a recommendation engine using R, leveraging topic modeling methods to increase user engagement by 85% by recommending academic journals based on browsing history.· Designed and developed a topic modeling framework using the R programming language· Utilized R libraries such as gensim, tm, and LDAvis to implement Latent Dirichlet Allocation (LDA) and other topic modeling algorithms· Preprocessed large text datasets, including cleaning, tokenization, and stemming, using R packages such as stringr, SnowballC, and quanteda· Conducted exploratory data analysis to determine the optimal number of topics and the most relevant terms for each topic· Built and validated the topic model, visualizing the results through interactive plots and graphs using the Shiny web frameworkProvided insights into the underlying themes and topics within the text data, enabling organizations to make data-driven decisions based on the extracted knowledge.

Designed a job recommendation system that pairs lexical features (using LDA) extracted from resumes with job postings. Conducted an experiment to validate a proprietary scoring algorithm for matching job seekers to employers.· Designed the experiment to identify differences between human classification and machine learning classifier· Hired recruiters to rate resumes to compare with machine learning algorithm ratings· Analyzed variance in scores by recruiters and algorithm to determine the consistency of the ranking system

Engineered innovative enterprise software module for automatic, optimal data visualization.· Developed advanced BI tools for data visualization and analysis, utilizing machine learning algorithms to automatically select and optimize chart types.· Utilized C++ and Java to analyze data and identify patterns, leveraging MPI libraries such as Boost.MPI and OpenMPI to improve performance and speed.· Utilized C++ STL and Boost C++ libraries for data structures and functionalities.· Implemented data visualization principles to create interactive and visually appealing charts and dashboards.· Used natural language processing techniques to understand the context of the data and the user's intent to determine the best chart type.

San Francisco Bay Area

Developed an AI-based system to classify 1.5 million food items in conjunction with USDA guidelines into three user-friendly categories: Red, Yellow, and Green. · Contributed to pre-formative phase of company by collaborating with CEO, my next-door neighbor, to apply machine learning and analytics to enhance health and wellness, resulting in the creation of a personalized nutrition platform and a company currently valued at $500 million.· Technical contributions included developing a topic modeling framework, written in R, to analyze food reviews, recipes, and articles.· Created a set of algorithms leveraging a variety of techniques, including multinomial logistic regression, support vector machine, and random forest to classify food ingredients as "healthy," "neutral," or "unhealthy,” leveraging data sets from the Food & Drug Administration· Collected and cleaned data from food databases, surveys, and records using R.

Developed an ML-based solution to ensure legal compliance of organizations' electronic communications· Designed and implemented a legal privilege determination system to comply with different jurisdictions' regulations.· Utilized JAVA programming language to build the system and Perl to perform data analysis.· Gathered legal requirements by reviewing relevant laws, regulations, and case law.· Developed algorithms to classify electronic communications as privileged or non-privileged.· Trained the algorithms using large datasets of labeled electronic communications.· Validated the system using a separate dataset to ensure accurate privilege determination.· Incorporated the system into the platform and continuously monitored its performance.

Engineered a recommendation engine in PLSQL, which sold to Facebook for $83 Million.• Derived a Bayesian filtering algorithm based on conditional entropy to remove low-signal content from the large-scale Facebook feed, while ensuring linearly efficient big O complexity, and personalization.• Ported algorithm from PL/SQL to JAVA for Facebook's recommendation engine, optimizingscalability, performance, user engagement, and revenue.

Implemented an intelligent pricing algorithm to optimize revenue for online ticket sales, yielding a 700X improvement in ticket sales.• Developed an adaptive ticket pricing system that optimizes revenue in real-time, resulting in a 30%increase in gross margin and ensuring events sell out at profitable prices.• Applied nonlinear dynamics, feedback control theory, data mining, and optimization to determinerevenue-maximizing prices for each ticket, resulting in a patented ticket pricing methodology.• Collaborated with the engineering team to seamlessly integrate the adaptive pricing system into aproduction environment.• Continuously optimized pricing strategies and increased revenue by monitoring and evaluating systemperformance through A/B testing and multivariate analysis.

Developed the System Dynamics-based, predictive, software simulator to dynamically forecast operating cost as a function of various macro-economic, and micro- economic variables.

Co-developed a software product to adaptively change a targeted email message within an email campaign, by dynamically measuring the level of recipient engagement.

Created the System Dynamics models, causal loop diagrams, and corresponding System Dynamics system of differential equations, for the IBM-Portland Smarter Cities Initiative

· Constructed forecasts to predict long-term Federal Reserve Bank prime rates.· Analyzed the impact of price changes on demand for Wells Fargo interest-bearing accounts. · Determined the optimal interest rate to pay Wells Fargo customers, minimizing account closures and CD tenure rollovers. · Conducted a variance analysis using SAS and SQL to assess yield dispersion among individual customers. · Conducted in-depth analysis to identify marketable population using customer profiles, credit risk, and segmentation data.

· Served as an internal consultant on key business initiatives, assessing the economic impacts of process improvements.· Developed a dynamic marketing attribution model based on the Markov chain algorithm as an alternative to traditional methodology (static, linear, rules-based attribution)· Developed a sales lead grading model to determine the probability of purchase for prospective Oracle customers. The program improved sales yields by 25% over the previous lead-scoring algorithm.