Reticular

What do they actually do

Reticular builds research-grade tools to inspect and steer protein language/structure models. They publish interactive demos and open-source code that visualize sparse, human-interpretable features inside models and show how nudging those features changes predicted protein structure or properties demo.reticular.ai, sae.reticular.ai, GitHub SAE repo, research blog.

Today they work through pilots and design‑partner projects with startups, pharma groups, and academic labs. Typical engagements take an existing model or embeddings (e.g., ESM), run Reticular’s interpretability pipeline, identify features that correlate with biological properties, and deliver visualizations and proposed steering interventions; wet‑lab validation remains with the partner. Evidence of active pilots includes public demos, contact/demo flows and TOS on their site, YC outreach for design partners, and a Lightning AI case study describing faster experimentation using their tooling homepage, YC profile, Terms of Service, Lightning AI case study.

Who are their target customer(s)

• Pharma R&D / early‑discovery teams: They must decide which targets and experiments to fund but can’t justify costly wet‑lab work on opaque model outputs; they need interpretable signals and curated validation cohorts to reduce decision risk.
• Biotech startups building protein design pipelines: They rely on off‑the‑shelf LMs/structure models but struggle to steer them toward desired properties because internals are a black box; they need to see which features control outputs and how to change them.
• Computational biology / ML teams inside pharma or biotech: They need reproducible, research‑grade methods to inspect and intervene in model internals so model‑driven design can be integrated into production pipelines without unexpected failures.
• Academic labs and protein researchers: They want visual, explainable mappings from model features to biological concepts to generate testable hypotheses and prioritize experiments, rather than chasing spurious artifacts.
• CROs and experimental partners: They need clearer, actionable model guidance (e.g., which sequence edits to test) to improve throughput and reduce wasted bench time on low‑confidence suggestions.

How would they acquire their first 10, 50, and 100 customers

• First 10: Run targeted, paid pilots via YC/advisor intros and inbound demo requests, each with clear deliverables (visualizations, 1–2 steering interventions, written validation plan) and a short joint technical note to serve as reference material.
• First 50: Package the pilot playbook (scope, timeline, price, success criteria) and run systematic outbound to biotech startups, mid‑size pharma computational teams, and CROs; add lightweight channels (conference demos, a small CRO/ML‑tools partner program) and publish case studies.
• First 100: Productize the pilot as a hosted “interpretability review” that runs on customer embeddings with standard dashboards and recommended edits; launch a self‑serve beta for repeatable users while keeping sales‑engineered pharma deals, expand integrations/partners, and add CS for onboarding and validation.

What is the rough total addressable market

Top-down context:

Global pharma/biopharma R&D is roughly $276B, with preclinical/early‑discovery representing about 30–43% (~$80–$120B) — the long‑term pool Reticular aims to influence by de‑risking target selection and early design (RD World; CBO; EFPIA).

Bottom-up calculation:

Near‑term, assume ~1,500 likely early adopters (computational teams across pharmas/biotechs/CROs) out of ~6,800 companies with active pipelines, each spending ~$100k–$250k annually on interpretability/steering tools and pilots — implying a ~$150M–$375M SAM, consistent with a broader AI‑in‑drug‑discovery market estimated at ~$1–$4B today (Statista; Grand View Research; GMI).

Assumptions:

• Only a subset (~20–25%) of organizations with active pipelines are near‑term buyers (have computational teams and relevant workflows).
• Average annual contract value ranges from ~$100k–$250k across pilots and early platform seats.
• CROs and large labs are included as additional buyers beyond the ~6.8k pharma/biopharma companies.

Who are some of their notable competitors

• Cradle: SaaS platform for protein engineering using generative models; competes for budgets where teams want end‑to‑end design, while Reticular focuses on interpretability and model steering.
• Profluent: LLM‑driven protein design; adjacent because customers choosing opaque generative outputs may also want interpretable controls and rationale for edits.
• Generate:Biomedicines: Industrial protein design company using ML; not a tools vendor, but targets similar decision points in discovery where interpretability could be a differentiator.
• BigHat Biosciences: Antibody/protein optimization with ML‑guided wet‑lab loops; serves similar optimization needs, though with integrated lab capabilities rather than interpretability tooling.
• LabGenius: ML‑driven protein engineering platform; competes for discovery budgets focused on protein optimization, while Reticular offers interpretable model controls rather than a closed pipeline.