Back to Nexomeric

Project Zero

AI-powered scientific research, starting with olfactory chemistry.

Project Zero is Nexomeric's first domain-specific research system – demonstrating our modular AI architecture through molecular analysis, property prediction, safety evaluation, evidence retrieval, and candidate exploration in olfactory chemistry.

Domain: olfactory chemistry
Stage: MVP in development
Architecture proof point

Product scope

The first domain for Nexomeric's scientific AI platform.

Olfactory chemistry is scientifically challenging enough to validate the architecture, narrow enough for an MVP, and commercially relevant – demonstrating how AI can help researchers navigate complex scientific design spaces before costly experimentation.

Molecular analysis

Validate SMILES, compute descriptors, compare structures, and build molecule profiles for olfactory use cases.

Odor prediction

Use olfactory models and evidence retrieval to estimate descriptors, odor-space similarity, and confidence.

Safety screening

Evaluate toxicity, structural alerts, regulatory references, and caveats before candidates move toward lab testing.

Evidence retrieval

Search scientific literature, patents, datasets, safety records, and similar molecules for grounded context.

Candidate exploration

Support inverse design workflows such as finding safer, cheaper, or more sustainable aroma molecule candidates.

Traceable outputs

Produce reports, rankings, next experiments, limitations, and workflow traces that researchers can inspect.

How it works

Planner first, tool-grounded at every reality checkpoint.

Project Zero separates reasoning from scientific computation. The planner decides what needs to happen, while specialist agents and tools produce the evidence.

Principle Engine

Loads olfactory rules, safety boundaries, task constraints, and reward criteria before execution.

Hybrid MCTS Planner

Explores candidate research paths, then executes reality checkpoints with real model and retrieval results.

Specialist agents

Coordinates retrieval, chemistry analysis, olfactory prediction, literature, validation, design, and insight synthesis.

Tool and evidence layer

Connects RDKit, vector search, knowledge graphs, olfactory models, chemistry models, patents, papers, and safety data.

Self-play debate

Advocate, critic, and judge agents can stress-test high-risk hypotheses, weak evidence, and low-confidence outputs.

Trajectory logger

Records tool calls, model outputs, limitations, estimated versus real rewards, and branches discarded or regenerated.

Representative workflows

From research questions to decision-ready outputs.

Find a cheaper alternative to a known aroma molecule
Assess cosmetic or personal-care safety risk
Explain formulation notes that differ from expectation
Explore novel musk-like or sandalwood-like candidates
Compare candidate molecules for R&D decisions
Help students map open questions in olfactory science

Who it serves

Built for chemical olfaction R&D teams and researchers.

Fragrance and flavor R&D
Cosmetics and personal care
Food and beverage aroma research
Pharmaceutical and aromatherapy teams
Home care and cleaning products
Industrial and environmental odor management

Validation plan

Honest about the hard science, explicit about how it will be tested.

Structure-odor prediction has sparse datasets, noisy labels, uncertain receptor-level biology, and out-of-distribution risk. Project Zero is designed to expose confidence and caveats while the MVP is validated.

Dataset benchmarks

Evaluate selected prediction workflows on GoodScents and Pyrfume with accuracy, F1, and cross-validation consistency.

OOD awareness

Test whether the system flags structurally novel molecules and poorly supported predictions before overconfident output.

Pipeline reliability

Compare multi-tool execution against single-model baselines for hallucinated citations, routing errors, and inconsistent results.

Expert review

Run structured evaluation with fragrance industry chemists or academic olfaction researchers once the MVP workflows are ready.

Current build

The MVP combines scientific data, model services, retrieval, and workflow visibility.

The first implementation is being built to prove whether the Nexomeric architecture can support useful olfactory chemistry workflows before expanding to adjacent scientific domains.

Python, FastAPI, and LangGraph orchestration
PostgreSQL, Qdrant or FAISS, and Neo4j evidence infrastructure
RDKit molecular processing and descriptor generation
Pyrfume and GoodScents data normalization
Model gateway for chemistry, olfactory, and safety services
React, Next.js, React Flow, D3.js, and planned WebSocket progress streaming

What comes next

One domain proves the architecture. The platform expands across science.

Once Project Zero validates the architecture, Nexomeric will expand across chemistry, materials science, biology, energy systems, climate technology, manufacturing, and other scientific domains – helping researchers accelerate R&D wherever experiments are expensive and design spaces are vast.

View roadmap