An agentic AI platform for generating IEEE-publishable, PRISMA 2020-compliant systematic literature reviews
IAMB the one.
iamb (n.): two beats, a soft one, then a strong one. 0, then 1.
I just got published in IEEE Access with a PRISMA systematic review. Reviews like that usually take about two years. IAMB.AI is built from the exact process that produced mine, rebuilt so AI does the grind, every step stays on the record, and you are the one who decides.
Invite-only while I dogfood it. The research cohort goes first.
The paper
Peer reviewed, published, and the reason this tool exists.
A Systematic Review of TinyML Techniques for Efficient Inference and On-Device Learning
IEEE Access, 2026
With Irin Jose, Nickolas Kenyeres and Kailey Cozart Quesada
DOI 10.1109/ACCESS.2026.3736856
TinyML is machine learning on microcontrollers: devices that cost a few dollars and run on almost no power. The review maps the techniques that make inference efficient on hardware that small, and the ones that let a model keep learning on the device itself.
Every stage of that review, from the protocol to the PRISMA checklist, is a stage in IAMB.AI. Same rigor, a fraction of the time.
How it works
IAMB.AI is an agentic AI platform for generating IEEE-publishable, PRISMA 2020-compliant systematic literature reviews. One road from a question to a paper. You always see where your review stands and what comes next.
- Question Write the question and the eligibility criteria.
- Search Search the open databases and IEEE Xplore from here, import the rest, log every search.
- Screen Two reviewers vote blind, human or AI.
- Full text Find the PDFs and assess every report.
- Extract Pull the data, with the quote it came from.
- Write Draft each section from your data, in your style.
- Check The PRISMA 2020 flow diagram and checklist, from the records.
- Publish A paper in the IEEE format, every page checked.
PRISMA 2020, all the way
The flow diagram is computed from the records, so its numbers always add up, and the 27-item checklist says where each item is reported.
AI reviewers that show their work
AI screens as an independent reviewer, argues its side in a conflict and extracts with quotes. Every judgement is stored with its model, prompt and rationale. People decide.
The paper at the end
The manuscript is written from the review's own data and built into a PDF in the IEEE format. Not a report about the review. The paper.
Before the review
BabySLR™
See your review before you write it.
A BabySLR™ is a small systematic literature review, run in one sitting, that the full one grows from. Before you commit months to a review, find out what the field has, where the gaps are and how big the review will be.
- Searches Your search string, in the real databases, through their official APIs.
- Screens An AI reviewer reads titles and abstracts, and the counts reconcile.
- Reads The likely studies: abstracts, and open full texts where they exist.
- Extracts Into your own data collection fields, marked preliminary for the team to correct.
- Runs the numbers A matrix of methods by outcomes, gaps from its empty cells, a meta-analysis when the numbers allow it.
- Synthesizes What the field says, with every claim traced to the study it came from.
- Feeds the proposal It becomes Appendix C, and the proposal argues from it.
It is preliminary on purpose: a sample, read mostly from abstracts, by AI, before the team has checked it. It steers the review. It is not a finding, and every view of it says so.
The plan
- research about researching research
- do research on ai research
- let the ai research the ai research
- profit
Straight from my bio. Step 0 is already running.
Research on research
AI research is the future, and it comes at two speeds.
AI-assisted research
AI reads, screens and drafts. People check and decide. This is where IAMB.AI works today.
Fully autonomous, agentic AI research
Agents run the whole review, end to end. I am building toward it without giving up the rigor of the first.
IAMB.AI is a dogfood tool. I use it to publish a lot. I may share it, the research cohort first, and let the quality of the papers make the argument.
Every AI step is logged with its model, prompt and outcome, so the process itself becomes data. Studying how the tool does research is part of the work.
How we work
A few rules we hold ourselves to, in our papers, in our code and in everything else we write, job posts included.
-
The business of research is social, and the paper is the product. A peer-reviewed paper is the hardest thing in our field to earn, and it is what we exist to produce. It takes a team that searches, screens, argues, agrees and writes together, so cohorts, rooms, boards and profiles make this a social network on purpose. We count papers accepted, not features shipped.
-
See #1. Seriously. Rule two is rule one again, because nothing we do matters more. That is serious play: the joke is light, the point is not.
-
Everything has a source. Our papers cite, and so does everything else we write: docs, decisions, even job posts. If we say it, we can show where it came from.
-
On the record. Every AI step, decision and change keeps who did it and why. Meetings end as records, not memories, and the knowledge in them stays findable.
-
AI does the grind, people make the calls. AI searches, screens, extracts, drafts and checks. People decide, verify and sign. AI is never an author.
-
Async first. Write it down. Five minutes of reading beats an hour of video, and what a meeting decides lands on the board as tickets.
-
Research on research. We study how research gets done, our own included, and change the process when the data says so.
-
We use it first. We publish with our own tools before anyone else does. If it is not good enough for our papers, it is not good enough for yours.
-
Quality is the argument. Rigor you can defend: PRISMA from the protocol to the checklist, every number traceable to its records.
-
Success is the one true goal. Papers accepted, people served, a company that lasts. We ask what the next step is, take it, and check whether it moved us. Our AI plans the same way.
-
Knowledge compounds. Every AI teammate keeps a memory of its work and shares it when asked, so the company remembers what any one of us would forget.
-
Innovate. Integrate. Iterate. Try the new idea, build it into the one system instead of beside it, and keep improving it until the numbers say it works.
-
Discover, not dictate. We find people, AI teammates included, the way life does: through real work and real choices. Hiring an AI teammate through the whole process gives different answers than writing it all down up front, and usually the right ones. This principle was found exactly that way.
-
Serious play. We make hard work feel like a game without making it any less serious. The stakes, the rigor and the paper are real; the play is how we keep going.
-
One human, a million AI. Every copy of this platform is a company: one person holds the offices and makes the calls, and a million-strong march of AI teammates does the work under their eye. That is how we build a billion-dollar company.
About me
ISRAEL CHEN received a Bachelor of Science in Computer Science, Bachelor of Applied Science in Supervision and Management, Master of Science in Software Engineering, and Master of Business Administration. He is currently pursuing a Master of Science in Computer Science with a specialization in artificial intelligence at the Georgia Institute of Technology, Atlanta, GA, USA.
Be the one.
IAMB.AI is invite-only while I dogfood it. Tell me what you want to research; the research cohort goes first.