Technology
We’re building a complete Agentic AI Platform for Biopharma — a system that not only generates outputs, but plans, reasons, and even writes their own code to pursue drug development goals. It includes Agentic Business Layer and Scientific Modeling Layer both rooted in high quality Data Infrastructure layer.
BIOPTIC AI agents as digital employees
BIOPTIC AI agents act as digital employees, able to think like biologists and medicinal chemists, pharmacologists and clinicians, BD analysts and supply chain specialists, health economists and market strategists — all at once. They don’t sit idly by, waiting for you to tell them what to do. They work autonomously, continuously monitoring a vast and diverse data landscape and bringing back critical insights.
Asset Scouting
Scans every paper, filing and news source in every language worldwide to surface and cluster deal opportunities.
Due Diligence
Assembles red‑flag memos from trial data, patents, and competitive intel in hours
Quality control
Benchmark-proven multi-agent QA with in-house checks, citations, and numeric plot audits.
Reinforcement loop
Continuous improvement through regular RLHF retraining that incorporates user queries and human feedback
BIOPTIC Scientific Models that ground our AI agents in real-world science
BIOPTIC Scientific Models have already discovered a number of novel molecules across multiple targets. Our Ligand-Based Model B1 makes screening ultra-large libraries almost unthinkably cheap. Our Structure-Based Model B2 outperforms AlphaFold3 and allows us to bypass the protein structure step, taking only the protein sequence and a SMILES string and outputting accurate activity scores. That means ~1700 molecules/sec on a single NVIDIA T4 GPU. Our award winning ADME/Tox Model B3 predicts molecular properties with
Proprietary Scientific Models
B1
Optimized for generalizability (i.e.scaffold hopping), excels
- on targets with known binders,
- screens ultra-large libraries
- at fractional cost.
B2
Predicts binders for novel targets.
- Rivals AlphaFold 3 and Boltz-2 for small molecule binding predictie
- No folding—protein sequence + SMILES → activity score
- Validated in-silico with SOTA benchmark PLINDER and iteratively tested in wet-lab
- ~1700 mol/s on one T4, sequence-length agnostic
B3
ADMET prediction
- Predicts ADME/Tox across diverse chemical space
- Benchmark-validated on novel scaffolds for broad transferability
- High-throughput: flags liabilities at the earliest design stage
Engineering Confidence at Every Byte
We sweat the plumbing so you can trust the output. Our agents pull only high-signal data, self-check it against human labels, and prove their work with public benchmarks. The entire stack runs firewalled and read-only by default, giving you rock-solid insight without handing your data to the internet.
Curated data stack
- To make sure only high-quality data is involved in critical decisions, we make LLMs self-reflect at every point of the deep research based on human-labeled data
- We fine-tune our agents to use trusted sources only: SEC filings, FDA guidelines, trials, publications from high-impact journals, market and media feeds
- Molecular models are trained on omics data generated in wetlab and licensed from our partners
Benchmarking
- We’re crazy about benchmarking
- We won two international competitions with 2000+ participants in Machine Learning for drug discovery and got top-2/3 on the ADMET properties prediction challenge with an innovative LLM-based approach
- We’ve built PLUMBER, the molecular binding affinity dataset for critical assessment of binding affinity prediction AI models
Data Security
- On-prem / VPC deploy – the whole stack runs inside your firewall; only a secure proxy reaches out to licensed feeds.
- Trusted-source tunnel – outbound traffic is limited to whitelisted IPs for trusted data sources; nothing else leaves.
- No auto fine-tuning – base weights stay read-only unless you approve a private fork.
Get in touch
Whether you’re a researcher, potential partner, or just curious about what we’re building, drop us a message. Explore how we can push the boundaries of science and discovery.