# Synthetic Sciences

> Synthetic Sciences is a Y Combinator (W26) startup building AI co-scientists that run the full research loop - literature review, hypothesis generation, experiment design, code execution on GPUs, analysis, and publication-ready drafts. Founded in 2025 by Aayam Bansal and Ishaan Gangwani, the company pairs a research product (synsci, plus the open-source OpenScience workbench and the Atlas compute layer) with a thesis that proprietary research-process data is the path to scientific foundation models with genuine research 'taste'. Its biology mode reached 92% on the BixBench Verified benchmark.

- **Founded:** 2025
- **Headquarters:** San Francisco Bay Area, California, United States
- **Founders:** Aayam Bansal (Co-Founder & CEO), Ishaan Gangwani (Co-Founder)
- **Team size:** 2
- **Products:** synsci (Synthetic Sciences platform), OpenScience, Atlas, Flywheel Mode
- **Notable:** State-of-the-art 92% on the BixBench Verified computational-biology benchmark., Accepted into Y Combinator's Winter 2026 batch., Raised a $1.4M pre-seed round.

## Products & services

- **synsci (Synthetic Sciences platform)** — AI co-scientists that take a research question, repo, or dataset and autonomously search and synthesize literature, generate hypothesis and idea trees, design experiment plans and GPU job specs, write and run code in containerized environments, launch and monitor experiments on serverless GPUs and clusters, and output publication-ready LaTeX drafts, figures, and slides. Operates in Research, Biology, Flywheel, and Write modes.
- **OpenScience** — Open-source (Apache 2.0), model-agnostic AI research workbench for ML, biology, physics, and chemistry. Browser-based workspace with a local agent runtime, 250+ editable skills, and direct access to ~30 scientific databases (UniProt, PDB, Ensembl, ChEMBL, PubChem, arXiv, OpenAlex). Works with any frontier or open-weight model.
- **Atlas** — Optional managed layer for agent-native research: a persistent project graph the work lives in, a library that grounds agents in live sources, managed multi-provider compute to run experiments at scale, and a state-of-the-art agent harness.
- **Flywheel Mode** — Post-training workflow that turns a team's own corrections, accept/reject signals, traces, and outcomes into training data, then handles the data plumbing, evals, deployment, and compute across 20+ providers so teams can fine-tune and own models instead of renting intelligence by the token.

## Achievements

- State-of-the-art 92% on the BixBench Verified computational-biology benchmark.
- Accepted into Y Combinator's Winter 2026 batch.
- Raised a $1.4M pre-seed round.
- Released OpenScience as a fully open-source (Apache 2.0) scientific AI workbench in July 2026.
- Founders have published together at NeurIPS, ICML, ICLR, AAAI, and CVPR workshops.

## Latest updates

- **2026-07** — Released OpenScience, an open-source (Apache 2.0), model-agnostic AI research workbench for ML, biology, physics, and chemistry.
- **2026-02** — Launched publicly on Y Combinator as 'AI Co-Scientists for End-to-End Scientific Research' with a launch video and $5 free sign-up credits.
- **2025** — Raised $1.4M pre-seed (as InkVell) and joined YC's W26 batch.

## Links

- Website: https://syntheticsciences.ai/
- LinkedIn: https://www.linkedin.com/company/synsci
- Twitter/X: https://x.com/SynScience
- GitHub: https://github.com/synthetic-sciences
- YouTube: https://www.youtube.com/watch?v=uHOmeCnV1HU

---

Profile page: https://yespress.io/synthetic-sciences-yc-w26
Published by YesPress — https://yespress.io
Last updated: 2026-07-30