Rivulet

pre-seed · 2026

                                                                                                                                                                                   
                                                                                                                                                                                   
                                                                                                                                                                                   
                                                                                                                                                                                   
                 ..                                .','.                                ',,                                ';i;'                               .:i;.               
               '...'                              .;,.';.                             '+;:;+'                             ,Oi,;O,                             :@0i0@:              
              '.   .'                        ...  ,:   :,  ...                        i;   ;+                        ','  i0   0i  ','                        #0   o@              
         .'.. .'   '. ..'.                  ::,::' ::,:: .::,::.                .:i;, ;+   i; ,;i;'                .o0+o0:.o0+o0.,0oi0o'                ,oO0; o#   #0 ;0Oo,        
        '.  '. '. .' .'  .'                ,:   ':,.....,;'   ::               '+:.'+:.+, ,+.:+'.:+'               +0   :0+'',''iO:   oo               :@+':@o.@i i@'+@:'i@:       
       .'   '' '. .' .'   '.               ,:.    ,:;:::,    .:,               ,+.  ii +; :+ ;i  .+:               i0'   .i0O000i.   '0i               i@'  #O @o +@ 0#  '@i       
        .'''' .'   '. ''''.                 ':;::'       ',:;:'                 :+i+i.,+' .+,.i+i+;                 ;0O0o;.  .  .;+0O0;                 o@O#O';@, '@i.O@O@o        
             .'.    '.                        ...,;'   .:,...                     ..';+,   '+;'..                     '',+O:   ,0+,''                     '',o@;   :@0:,'          
        .''''.       ..'''.                 ,:::: ';. .;' ::::,                 :i++i:.      ,;i+i:                 i0o0o.:O' 'O;.o0o0i                 i#@#Oi.  .  .i0#@#+        
       .'     .'''''.     '.               ';.  :: :, ,; ,:   ;'               :+'    :++i++:.   .+;               :O'  oo 0i iO +0  .O;               +@,   .+#@#@#+'   ,@o       
       .'   .'.     .'.   ''               .;'..:, ;' '; ';..';.               ;+   ,+;..'..:+,   ii               ,O;',0i.O; :O.;O,';O,               0#   ;@o',:,'o@;   OO       
        ''.''. ''.''  ''.''                 .,:,' ,:   :, ',:,.                .i+;i+,.ii;ii.'+i;ii.                'io+: i0   0+ :+oi'                '##oO@;.##o##':@Oo##'       
              .'   '.                             :,   ,:                        .''  :+   +;  ''.                        0+   i0                        ':,  +@   #o  ,:'         
               '. .'                              .;,',;.                             '+:':+'                             ,O+;+O,                             :@+:+@:              
                ...                                 .''                                .:;:.                                ,::.                               '+o+,

// AGENT MODE — STRUCTURED DATA VIEW // or press V to return

Slide 01 — The physical layer for AI drug discovery

Agent-Executed Biology. Model-Ready Data.

The physical layer AI drug discovery runs on.

Rivulet

Asmay Gharia, PhD · Founder & CEO

asmay@rivulet.bio · rivulet.bio

01 / 08

›

Slide 02 — The Gap · Agents run experiments on hardware designed for humans

Rivulet

02 / 08

Agents read. Agents design. Agents plan. Then the loop hits hardware built for human operators.

AI at Machine Speed

167M

cells · Arc State observational training set

225K

perturbations · Tahoe × Arc × Biohub partnership scale

$4B+

committed · Xaira, Isomorphic, Insitro

Built for Human Operators

0.01%

of AI-generated candidates reach physical testing · results arrive as a file when the run ends

24–72h

biological incubation floor per run · the agent gets no signal until the run ends

16–40 wk

campaign from HTS setup to validated hits · serial by design · no concurrent runs

The agent runs. The hardware makes it wait.

‹

›

Slide 03 — Solution · Chip designed for AI native execution

Rivulet

03 / 08

Native hardware for agentic drug discovery. Loop closed.

The first chip architecture to run biology in parallel, read it continuously, and stream model-ready results on demand. Agent-callable.

PEDOT:PSS conducting-polymer electrodes move individual cells and droplets using dielectrophoresis. A proprietary ML classifier reads each particle's impedance signature live. Every run emits a flat, per-particle interaction table: the exact shape foundation models consume.

10⁶

observations per run

5 pL

minimum control volume

sub-ms

per-event ML inference

API

agent-callable · MCP + CLI

Device breakdown → Appendix A01

‹

›

Slide 04 — Demo · Claude called our chip

Rivulet

04 / 08

Demo

Claude called our chip.

01 Claude receives the prompt

02 rivulet-mcp → ExperimentPlan

03 Chip routes, mixes, reads live

04 Panels update in real time

05 Agent receives model-ready data.

Any agent. Any experiment. Model-ready output, on demand.

Access the MCP server today at github.com/AsmayGharia/rivulet-mcp-public

‹

›

Slide 05 — Traction · NIH pilots, NCI-validated performance, pharma AI-discovery LOI

Rivulet

05 / 08

NIH pilots. Validated performance. Pharma AI-discovery LOI.

Therapy target reached pre-clinical viability · NCI · solid tumor program

10x+

Real-time selection gain · 49% vs. 4.7% standard · NCI myeloid program

Active Pilots: NCI · NIAID

"This technology made me rethink what was possible with cell therapy."

— Dr. James Cronk, MD, PhD · Cincinnati Children's Hospital

Active Pilots

Pipeline

Top 10 pharma

Full validation data → Appendix A02

‹

›

Slide 06 — Market · Data gen for AI-bio labs first, pharma next, agent infrastructure at scale

Rivulet

06 / 08

Data gen for AI-bio labs first. Pharma next. Agent infrastructure at scale.

The product is model-ready biology data for AI drug discovery labs. Customer count × access fee, in three waves.

Year 3

Data Gen for AI-Bio Labs

$15M ARR

15 accounts × avg $1M ACV. Mix of pilots ($600K) and first production deployments ($1.8M). Pilot-to-production starting from 3 active programs today.

Year 5

+ Pharma ML R&D

$60M ARR

50 active programs across 8–12 accounts × avg $1.2M ACV. Pharma buys per program: 3–5 accounts × 8–12 programs each. Lilly, Novartis, BMS, Pfizer, Genentech.

Year 7

+ Vertically Integrated Agent Layer

$150M+ ARR

Existing accounts compound: a program at $1.2M/yr reaches $4–5M/yr as Rivulet becomes the execution layer the agent calls directly. Usage scales with run volume. The chip moves from instrument to infrastructure.

10x Genomics reached ~$300M revenue by year 7 from founding. Same playbook: sell to the labs that need the hardware first.

‹

›

Slide 07 — Team

Rivulet

07 / 08

Built by the people who invented the hardware.

Asmay Gharia, PhD

Founder & CEO

Six years at Cambridge (EE PhD) inventing the hardware that became Rivulet. Exclusively licensed to Rivulet. First-author, Science Advances 2024.

asmay@rivulet.bio

Katherine Masih, MD, PhD

Co-Founder & CSO

Physician-Scientist Resident at Stanford. Cancer genomics and pediatric cell therapy, the domains behind Rivulet's NCI and Cincinnati Children's pilots. MD, U Miami; PhD, Cambridge.

Austin Weinstein, MBA

Co-Founder & COO

Cancer Entrepreneurship Fellow, Dartmouth Tuck. Owns GTM and commercial. Closed St. Jude Children's Research Hospital and Celdara end to end.

Scientific Advisors

Prof. George Malliaras

Scientific Advisor

Prince Philip Professor of Technology, Cambridge · Bioelectronics Laboratory · world expert on PEDOT:PSS · spun out 7 biotech companies

Dr. Iain Fraser

Scientific Advisor

Chief, Signaling Systems Section · NIAID, NIH. Leads the active Rivulet NIAID pilot program.

Bryan Poltilove

Industry Advisor

Former Head of Cell Therapy, Thermo Fisher · Former Partner, BroadOak Ventures. Advising pharma market entry and commercial partnerships.

‹

›

Slide 08 — Ask · $2M pre-seed

Rivulet

08 / 08

$2M to put the first chips in labs.

This round funds a deployable pilot testbed that de-risks the platform for early partners, demonstrates high-throughput potential, and lets them pre-integrate before Series A.

Use of Funds

Build and ship a deployable pilot testbed

Enable pilot partners to de-risk the platform hands-on

Demonstrate high-throughput capability

Pre-integration ready: agent-callable CLI + HTTP API + MCP server

Path to Series A

→ 1+ pharma deployment live (on-premise)

→ Chip at scale with contract manufacturing partner

→ Device engineer, biologist, AI engineer hired

Infrastructure-layer positioning: 3 paying pharma customers and a growing developer ecosystem.

Raised to date: $510K F&F

Compute needed for biology is infinite. Rivulet makes it tractable.

Join us in building the substrate vertical agents run on.

‹

›

Appendix gate — Deep dive materials across 4 sections

12 SLIDES · 4 SECTIONS

Deep dive materials

How it works · who buys it · why we win · timing.

← Return to Ask

Rivulet

‹

›

Technical

A01 — How It Works · Three published primitives on a single chip

Rivulet

A01 / 12

Three published primitives, integrated on a single chip.

Rivulet chip physics — DEP routing and impedance classification pipeline

Min volume: 5 pL · Electrode: PEDOT:PSS · Search scale: 10⁶ · MCP + CLI + HTTP API

01 PEDOT:PSS Electrodes

Conducting-polymer electrode layer enabling both DEP routing and impedance sensing in a single interface. The material that makes simultaneous manipulation and reading possible on one chip.

Gharia et al., Science Advances, 2024

02 Dielectrophoresis

Electric-field gradients move individual cells along deterministic waypoints. No pumps, no tubes, fully programmable.

Tian et al., Microsystems & Nanoengineering, 2024

03 ML Impedance Classification

Multi-frequency electrical fingerprint → cell type → cell state, label-free, mid-flight.

Kim et al., Sensors & Actuators B, 2025

‹

›

A02 — Validation Data · 49% editing efficiency — NCI myeloid, Lonza Nucleofector 4D comparator

Rivulet

A02 / 12

VALIDATION DATA

Per-cell precision crosses the threshold bulk hardware cannot.

This data is from an early alpha focused on one modality: myeloid cell therapy. In it, we derisked high-throughput single-cell precision and effectiveness above the clinical viability standard. The current device requires no calibration for the assay or task. Same chip architecture. Any biology.

Study: Second-generation myeloid cell therapy for brain tumors and metastatic solid tumors.

National Cancer Institute's Center for Cancer Research · 2025 · Rivulet: pre-clinical validation phase

Condition	Edited
Control	4.74%
Lonza Nucleofector 4D	5.25%
Rivulet	49.1%

Clinical viability threshold: ~20%

Ctrl

Lonza

Rivulet

This is not an incremental improvement. At 5.25%, the best commercial alternative barely outpaces the control and falls far short of clinical viability. At 49%, Rivulet crosses the threshold by 2.5x. The difference is per-cell precision.

‹

›

A03 — The Output · The data foundation models actually want

Rivulet

A03 / 12

THE OUTPUT

The data foundation models actually want.

Well-plate averages throw away the signal.

Arc's State, Tahoe's atlas, Recursion's maps all consume the same shape: flat, per-cell, per-perturbation rows with provenance. Rivulet preserves it by construction. This is what turns a model's prediction into a finding.

· One row per single-cell interaction
· Full provenance: chip → run → plan → waveform
· Drops into anndata / Parquet / your agent

CLI + HTTP API + MCP server.

run_4812.parquet

particle_id: 0x7fa2

class: "t_cell"

confidence: 0.94

counterparty: ["drug_A", "drug_B"]

reaction_zone: "rz_03"

outcome: "synergy"

duration_ms: 14.2

sort_lane: "lane_0"

plan_id: "plan_4812"

‹

›

Commercial

A04 — The Buyer · AI drug discovery labs blocked on validation hardware

Rivulet

A04 / 12

THE BUYER

AI drug discovery labs whose predictions are blocked on validation hardware.

AI Drug-Discovery Labs

Xaira · Isomorphic · Insitro · Recursion · Latent · Noetik. Foundation-model teams who need perturbation data at a scale well plates cannot produce. Model output exceeds wet-lab validation throughput by 100×.

Pharma ML R&D Groups

Lilly · Novartis · BMS · Pfizer · Genentech. Combinatorial screens where hit rate and synergy depend on single-cell resolution. Well plates can't resolve at single-cell scale.

Autonomous / Agentic Labs

Self-driving lab teams that need a physical endpoint an agent can actually call end-to-end. Already using AI for candidate generation. Frustrated that validation is the bottleneck. Currently in data exchange under NDA.

‹

›

A05 — Pricing · Three phases from pilot to infrastructure

Rivulet

A05 / 12

PRICING

Three phases from pilot to infrastructure.

Single monthly contract covers platform access. Volume scales with you.

Now → 18 months · Pilot Contracts

$50K / month

3–6 month engagement · ~$300–600K total

Cartridges + reagents included
Platform access on our hardware
Model-ready data delivered as output
Pilot success criteria agreed up front

18 months → 3 years · Production Deployments

$150K+ / month

Annual contract · ~$1.8–2.4M ACV · + per-interaction charges

On-premise Rivulet deployment · CLI + API + MCP
Priority support + SLA
Classifier fine-tuning on customer data (private)

3 years+ · Subscription + Usage Tail

Platform fee + per-interaction usage

Closed-loop AI discovery subscription

Token-style charges per interaction processed
Revenue scales with volume without proportional cost growth

‹

›

A06 — Roadmap · Where the $2M seed takes us in 18 months

Rivulet

A06 / 12

ROADMAP

Where the $2M seed takes us in 18 months.

Q2 '26

Alpha Platform · Now

Cartridges + dashboard; prototype shown.

Q4 '26

3 Paid Pilots Live · +6 mo

NCI myeloid + first pharma screen.

Q2 '27

First On-Prem Deploy · +12 mo

CLI + API + MCP; cell-state classifier shipped.

Q4 '27

Million-Scale Run · +18 mo

Protocol Runner v2 at pharma partner.

Series A thesis: Infrastructure-layer positioning: 3 paying pharma customers and a growing developer ecosystem. NIH SBIR Phase I: LOI submitted ($300K non-dilutive). Decision pending. Extends runway. Not a dependency.

‹

›

Competitive

A07 — Competition · No incumbent has the richness, scale, or FM-native featureset

Rivulet

A07 / 12

COMPETITION

No incumbent has the richness, scale, or FM-native featureset.

Company	What They Sell	Per-Cell Data?	Agent-Callable?	Why They Can't Catch Up in 24 Mo
Rivulet	Chip + data platform	Yes — flat rows, live	Yes — CLI, API, MCP	—
Bruker (Berkeley Lights / PhenomeX)	Beacon optofluidic platform	Per-pen, light-isolated	No	Throughput caps at ~10⁴ pens; $1M+ instrument tied to biologics workflows
Fluidic Sciences (Sphere Bio)	Picodroplet emulsion platform	Per-droplet, offline read	No	Cells locked in droplets; no live manipulation or per-cell sort
DropGenie	Digital microfluidics electroporation	Bulk per-droplet, no per-cell	No	Capped at 48 reactions per cartridge; transfection-only
dropXcell	Double-emulsion droplets + FACS	Per-droplet via FACS	No	FACS-dependent batch workflow; antibody-function screening only

Every run sharpens the classifier. Every deployment deepens the data moat.

‹

›

A08 — Moat · Data moat compounds — IP is published and licensed

Rivulet

A08 / 12

MOAT

Data moat compounds. IP is published and licensed.

DATA FLYWHEEL

Every run trains our classifier on more cell states. Better classifier → better S/N → smaller samples → more runs.

⚙

More Runs

◈

Better Classifier

≋

Better S/N Ratio

◎

Smaller Samples

IP & SCIENCE

Core particle manipulation platform

Exclusive license · granted · priority Mar 2022

DEP + PEDOT:PSS

Exclusive option · pending · filed Mar 2023

FTO

Positive initial analysis · full FTO Q3 2026

Peer Review

Gharia et al., Science Advances 10(43), 2024

Classifier

Proprietary ML · no cross-customer training

‹

›

Timing

A09 — Why Now · Three forces converged in 2024

Rivulet

A09 / 12

WHY NOW

The window is open now. Three forces converged in 2024.

↑

Foundation models outrun wet labs.

AI-native labs generate predictions faster than any pipette can validate. Arc shipped 170M cells. Tahoe shipped 225K perturbations. The bottleneck moved to hardware in 2024.

◉

The science is published and licensed.

PEDOT:PSS is peer-reviewed. The DEP platform is exclusively optioned from Cambridge. The hard technical risk is resolved. What remains is scale.

◈

A new buyer category appeared.

AI-native drug discovery labs spend at pharma scale and move at startup speed. Xaira, Isomorphic, Insitro, Recursion. They didn't exist as a category three years ago and have no incumbent vendor.

These three forces converged in 2024. The window is open now.

‹

›

A10 — What Can Kill Us · The three risks we think about every day

Rivulet

A10 / 12

WHAT CAN KILL US

The three risks we think about every day, and how we handle each.

01 AI-Bio Labs Build In-House

Why it happens

Xaira, Isomorphic, Insitro together hold $4B+ and could fund their own chip teams.

How we mitigate

We sell the shovels. Integration cost, licensed IP, and a 5-year head-start make buy > build. We are a platform, not a competitor to their core mission.

02 Pharma Adoption Cycle 24–36 Mo

Why it happens

R&D sales into Lilly or Novartis are long and political. A $2M seed relying on pharma by month 18 is fragile.

How we mitigate

Academic + AI-bio first (12-week pilots, process validation). Pharma pulled in via institutional logos. Two of three current pilots are outside pharma.

03 Chip Manufacturing Yield

Why it happens

Deep-tech startups die in the contract manufacturing handoff. Microfluidic yield can fall off at scale-up.

How we mitigate

PEDOT:PSS process is published (Science Advances 2024) and runs in third-party labs today. Two CM partners qualified pre-seed. IP covers high-yield variants.

‹

›

A11 — Validation · First commercial validation · Top 10 Pharma NDA detail

Rivulet

A11 / 12

VALIDATION

First Commercial Validation

Top 10 Pharma NDA

Signed with a top-10 pharma company. Active data exchange in progress. Scope: validation of DEP sorting on customer cell lines, impedance classifier benchmarking vs. existing flow cytometry. Identity available under NDA.

‹

›

A12 — Why Rivulet, Why Us · Founder origin story

Rivulet

A12 / 12

WHY RIVULET, WHY US

Massively parallel, validated biological data used to be impossible. Now, it’s necessary.

ASMAY GHARIA, PHD · FOUNDER & CEO

2018–2024 · Cambridge PhD

Six years inventing the hardware that became Rivulet. Advised by Prof. George Malliaras (Prince Philip Chair, Bioelectronics). First-author, Science Advances 2024.

2024 · The Inflection

By 2024, AI labs were generating predictions faster than any pipette could validate. Arc shipped State at 170M cells. Tahoe shipped 225K perturbations. Every frontier AI-bio team was still making validation data with pipettes. The discoveries weren’t landing. The hardware had to change.

2025→ · Rivulet

Incorporated with Katherine (MD, PhD, Stanford) and Austin (MBA). $510K F&F raised. Pilots live with NCI, NIAID, and a pharma partner under NDA.

‹

›