How Lens replaces
chance with evidence.

Six layers of agentic infrastructure, built on the world's largest real-world multimodal dataset.

Book a demo
The platform · layer by layer
06
Security
→
05
Scientific Workflows
→
04
Agents · Tempus One
→
03
AI Workspaces
→
02
Frontier AI Models
→
01
Multimodal Patient Data
→

The agentic AI platform purpose-built for drug development.

Fueled by the largest real-world multimodal dataset. AI agents that design, execute, and interpret scientific workflows — expanding what your team can do.

Layer 01 · Multimodal Patient Data

The world's largest real-world multimodal dataset. Built for AI.

45M+ patient cases. DNA, RNA, digital pathology, radiology imaging, and clinical outcomes linked at the patient level.

Most "AI for drug development" platforms assume the customer brings the data and the vendor wraps a model around it. Lens inverts that.

A decade of harmonization and provenance work sits underneath. Your team inherits all of it. Bring sponsor RWD or registries when you need to — everything plugs into the same stack.

AI-ready by default. Structured and unstructured records for foundation model training. Medications, diagnoses, demographics, biomarkers, surgical, and radiation histories, all extracted from unstructured data and ready for analysis.

Tempus data snapshot: more than 45 million total patient records, 8 million imaging records, 4 million samples sequenced, 350,000 DNA + RNA profiles
Domain-trained
Digital pathology
Biomarker prediction trained on millions of de-identified slides.
Domain-trained
Multiomic models
Fuses DNA, RNA, and clinical signals at the patient level.
Domain-trained
Outcome prediction
Grounded in patient-level real-world outcomes data.
+ Frontier LLMs
GPT · Claude · Gemini · Llama
Routed in for reasoning, language, code.
Layer 02 · Frontier AI Models

The right model for every task.

Frontier large language models plus domain-trained foundation models.

Lens is model-agnostic. Every query routes to the best model for the job. Frontier LLMs handle reasoning and code. Tempus's domain-trained multi-modal foundation models bring the scientific rigor.

Ask a biomarker question and Tempus One doesn't pick one model and call it done. It routes to the domain-trained model that knows the biology. Validates with a frontier model. Grounds the output in real data.

Model selection is transparent. Every output shows model selection, pipeline parameters and the underlying data.

Layer 03 · AI Workspaces

A workspace built for AI workflows.

Dedicated, secure cloud workspaces with pre-configured libraries and compute infrastructure. Zero setup.

Each sponsor gets a dedicated, isolated environment. A collaborative workspace with Python, R, Jupyter, SAS and VSCode.

Mount your own data directly into your Lens Workspace as if it were a local file system — alongside Tempus's 45M+ real-world dataset.

The workspace ships with what scientists actually use: multimodal data, notebooks, GPU-powered training and inference, version control. Compute scales with workload. No data reformatting. No engineering tickets to get started.

Tempus One co-scientist drafting an analysis plan: TRGT1 Expression and CRC 1L Outcomes
Layer 04 · Agents · Tempus One

Your co-scientist, in production.

Tempus One doesn't just answer questions about your data. It does the heavy lifting.

Ask in plain language. Watch a validated cohort assemble in minutes. Generate a feasibility analysis with a real-world control arm. Hand off a regulatory-grade evidence pack.

Tempus One is not an LLM wrapper. It's a production agent with native access to your data and models. Built so every role on your team stays in the loop and moves faster together.

01
"Our Phase 2 TROP2 ADC in 2L NSCLC — recommend an enrichment strategy: which expression cutoff, histology, and co-alterations give us the strongest effect size in the RWD cohort?"
Enrichment definition with subgroup sensitivities · 6 minutes
02
"Pressure-test the I/E for our HER2-low gastric Phase 2. Quantify the tradeoff of loosening prior-lines (2 → 3+) and ECOG (0–1 → 0–2): pool growth vs. projected ORR shift."
Criterion-level attrition, strict vs. relaxed comparison · 8 minutes
03
"Our FGFR2-fusion inhibitor is approved in cholangiocarcinoma. Where else do FGFR2 alterations co-occur with poor outcomes and unmet need — rank the top three."
Three ranked indications, biology and SoC evidence · 9 minutes
Layer 05 · Scientific Workflows

Every stage of drug development. Built in.

Five core workflows. End-to-end. Refined every quarter.

Lens ships with five pre-trained workflows. Each workflow provides a collaborative workspace where your team works alongside Tempus One to move from raw data to supportive evidence in a fraction of the time.

WORKFLOW 01
Target validation
Hypothesis to multimodal evidence to patient subpopulation. Cohort exported, ready for the next decision.
WORKFLOW 02
Biomarker discovery
Cross-indication signal exploration with whole-transcriptome data and patient-level outcomes.
WORKFLOW 03
Trial design
Plain-language protocol to validated patient cohort, with eligibility checks built in.
WORKFLOW 04
Trial feasibility
Real-world cohort, control arm, site list, and enrollment projection in minutes.
WORKFLOW 05
Evidence generation
Rigorous evidence packages with complete audit trails.
EXTENSIBLE
Custom workflows
New workflows ship every quarter. Customer-specific workflows on request.
Every workflow gets faster every time you use it. Lens carries your team's institutional knowledge forward, built on the work your team has already done.
Layer 06 · Security

Trust and data protection wrap the entire stack.

Essential security measures for drug development.

Data protection wraps each layer for security and control.

HIPAA compliant
All real-world data is de-identified under Safe Harbor or Expert Determination methodologies before reaching any model or agent.
End-to-end encryption
All data is encrypted in transit and at rest. Keys are managed inside the customer environment.
Role-based access (RBAC)
Permissions are scoped to role and program. Access is authenticated, authorized, and logged.
Zero AI training on sponsor data
Sponsor data is never used to train Tempus foundation models without your permission. Architectural commitment, contractually enforced at the infrastructure layer.
Questions

What buyers ask first.

How is Lens different from a real-world data database or other AI platforms?
Lens combines the largest multimodal real-world dataset with purpose-built AI tools that generate answers your team would have written queries to find. Our data and insights are proven to increase the probability of technical and regulatory success, and we've invested significantly to deliver them at 10x the speed for scientific teams at global pharma. The highest-quality data is table-stakes; the real differentiator is how fast you can convert that data into the critical decisions your drug development program needs.
What happens to my data when I bring it into Lens?
Sponsor data stays in your dedicated, secure environment. Tempus does not train models on it. You retain full ownership and control. Every agent action against your data is logged in an audit trail you can export.
How does Tempus One make decisions?
It doesn't. Your team does. Tempus One surfaces the optimal path given your data, your protocol, and your constraints. Every recommendation includes its sources, its confidence, and an audit trail. Scientists stay in the loop on every decision that matters.
Is Lens oncology-only?
Today, the deepest multimodal data is in oncology, which is where the foundation is strongest. Lens itself works across therapeutic areas, and the substrate is expanding accordingly.
How quickly can a team get started?
Standard onboarding is measured in weeks, not quarters. Book a demo and we'll walk through your specific program, typically with a translational lead and a procurement contact in the first call.
Get started

Replace chance with evidence.

30 minutes. Bring your program. We'll show you what Lens can do for you.

Book a demo
Book a demo →