Q: WHAT KIND OF AI ENGINEERING EXPERIENCE DOES MANIFOLD HAVE?
We have experience in a range of areas of AI. Our intelligent IoT practice provides expertise across devices, edge, cloud, vision, voice, language, sensors, time-series, machine learning, and signal processing. Our enterprise data practice provides expertise across cloud, data platform, data engineering, data pipeline, data profiling, data unification, NLP, natural language understanding, search, conversational AI, entity resolution, data science, and predictive analytics.
Manifold has worked with Global 500 and high-growth companies across multiple verticals, including consumer electronics, industrials, wireless, online commerce, and digital health. To read about specific examples of our previous client engagements, we welcome you to peruse our case studies.
Q: HOW DOES MANIFOLD STRUCTURE CLIENT ENGAGEMENTS?
We engage with our clients using a deliberate process. Projects generally include a kickoff phase, followed by two-week-long development sprints, and wrap up with a hand-off phase. During the kick-off, we will review this baseline process and modify it as needed to better fit your needs.
Our Lean AI process—reflecting decades of collective experience—enables building AI systems with minimal wasted effort and time. Lean AI combines key principles from the areas of software development (Agile), technology startups (Lean), and data science (CRISP-DM).
Q: HOW DOES MANIFOLD STAFF A TEAM FOR EACH CLIENT ENGAGEMENT?
The engagement team typically consists of an engagement manager and 2 to 6 engineers, depending on the specific scope. It's very difficult for one engineer to embody all the required skills across the spectrum from creating complex data pipelines to complex ML models, so our delivery teams include a mix of data engineers, data scientists, and machine learning engineers with the requisite skills as a team. This provides us with a balance of experts in platforms and architecture as well as models and algorithms. The team takes time at the beginning of a project to think ahead to deployment, and engineer a robust architecture before we jump into the data modeling so we can accomplish our goals and make the final shift into production as seamless as possible.
For some client engagements, Manifold may recommend including a specialized expert in a certain technology or industry. Our network of superlative advisors and collaborators allows us to partner with experts beyond our core team when needed.
Q: HOW DOES MANIFOLD COMMUNICATE WITH CLIENTS?
Manifold works in two-week-long sprint cycles, following the established agile development best practices. Our team meets to plan out what we need to accomplish at the beginning of each sprint, focuses our execution to that plan, and shares the results at the end of each sprint. Because data science and engineering are iterative processes, with learnings and modifications, we communicate closely with you during every sprint to ensure we remain tightly aligned and focused on the problems with the highest value to your business.
Typical communication touch points include a mix of milestone-based and ad-hoc discussions.
- Kickoff planning meeting
- Next-sprint planning meeting & sprint-end retrospective
- Week-in-review detailed email
- Clients always know what is happening via access to the project management system (Trello), chat room (Slack), version control system (GitHub), and ongoing communication with our teammates.
- Peer development with client engineering team in-person or via video chat
- Daily standups that the client engineering team can join
- Check-ins with business stakeholders by email, phone, or video
Q: WHERE DO MANIFOLD ENGINEERS WORK?
Manifold engineers primarily work out of our offices in Silicon Valley or Boston. Manifold team members are on-site at key milestones and, as needed, for paired development with our client's engineering team. We communicate frequently, regardless of where we are.
Q: HOW DOES MANIFOLD HANDLE SOFTWARE DELIVERABLES?
Manifold delivers software to our clients in a systematic way. Towards the end of the engagement, we take the following steps to ensure a smooth handoff:
All software is written in Python and executable within a Docker container. All of our source code is checked into a private GitHub repository in our account that is transferred to you at the end of the engagement.
We perform a high-level code walkthrough together to ensure that your engineering team has a solid understanding of what we have developed, and to answer any questions about how the project was structured.
We review with your team—or train them if they’re not already familiar—how to run Python Jupyter notebooks, start and run the Docker environment, and execute any command-line scripts.
We provide detailed training to your data scientists on the algorithms and models we have developed.
If you are planning to integrate our software directly into a production environment, we will set up a user acceptance testing (UAT) session and work closely with your operations team to ensure a smooth deployment.
We recommend you appoint a primary point of contact that will accept this delivery—typically a senior engineer or data scientist with whom we will work closely throughout the engagement.
Q: WHY DO CLIENTS WORK WITH MANIFOLD INSTEAD OF BUILDING A TEAM IN-HOUSE?
Sometimes our clients find themselves comparing this to hiring a team at the company. Working with us reduces risk substantially:
- We recruit 10x engineers, many of whom we've worked with previously in our careers.
- Project team members have worked together using our "way" of doing things.
- Defined projects require far less commitment than hiring full-time teams.
- We can quickly rotate in a new team member if someone is out unexpectedly.
To learn more about how we work with our clients and what we can do, see our Practice Areas.
Q: WHY DO CLIENTS WORK WITH MANIFOLD INSTEAD OF ONE OF MANIFOLD'S COMPETITORS?
Robust data products rather than more data analysts: A good heuristic for whether you're on a path to a serious long-term data roadmap is your answer to the question: "How many models—machine learning or otherwise—are running in production today?"
Some organizations have approached the need for being more data-driven by seeking out a large pool of data analysts, often from an offshore analytics staffing firm. However, we've found they often need help from a team like ours to clean up when they discover much too late that the work was either not done right or was not building towards a serious long-term data roadmap.
Instead, our approach is to replace the need for ever more data analysts by building a series of data products or what Professor Tom Davenport referred to in recent Harvard Business Review article as autonomous analytics.