Agentic engine · Clinical research · Colombia / Chile

Agents run the recruitment. People make the decisions.

Chain Reaction operates the clinical trial recruitment funnel end to end: it sources, pre-screens, supports, and documents. Every output is signed, and nothing moves forward without validation from the site team.

01 · The problem

The funnel leaks at both ends.

At the top, campaigns that pay for leads who will never qualify. At the bottom, patients who meet the criteria and still never enroll: out of fear, logistics, or slow response times.

≥ 50%At least 50% of qualified patients never enroll. The budget is spent at the top; the value is lost at the bottom.
Fig. 01 · Recruitment funnel · livetwo ends
LeadsPre-screeningconversationsMeetthe criteriaEnrolledend A · targetingbudget is spent hereend B · post-qualif.≥50% never enrollillustrative scale · each site's real percentages are measured in the baseline audit
02 · The work

Six fronts, and one governs the rest.

Each front runs one part of the journey and leaves a record of what it did. The first is not a stage of the funnel: it is the condition for the funnel to exist. They are described by their outcomes: the internal design is part of the protected value.

  1. The framework the site approves

    The study documents —protocol, informed consents, sponsor guidelines, the site’s own standards— are loaded and turned into concrete operating rules, each one carrying the verbatim quote from the document behind it. The site approves them one by one, with the quote alongside.→ until they are approved, the protocol takes no patients through any channel

  2. Compliant outreach

    Campaigns and outreach built to regulation from the start. The package that reaches the IRB / ethics committee arrives ready for review, not for objections.

  3. Conversational pre-screening

    Talks with patients over WhatsApp in plain language, captures the relevant clinical profile, and prioritizes candidates with a documented rationale.→ the coordinator validates every candidate before anything moves forward

  4. The tests that are missing

    The patient sends their test results through the same conversation: they are filed with their provenance, and the patient always gets confirmation of what happened. The platform reads what the document says and presents it with its quote and marked as pending confirmation, never as verified data.→ only a person on the team turns a reading into site data

  5. Post-qualification support

    Detects real barriers (transport, fear, scheduling) and maps them for the coordinator. It informs; it never persuades.→ any clinical question escalates to a person

  6. Reactivation and referral network

    Patients already screened return to the process when an active protocol matches them; physicians and institutions refer candidates with full traceability of where each one came from.→ reactivation, only with valid, current consent

03 · The limits

What the system is barred from doing.

A serious instrument is defined by its limits, too. These come from the design of the system, not from a manual of good intentions.

Register of prohibitionsby design · not by policy
  • Does not diagnose

    And it gives no medical advice. A clinical question is transferred to a person on the team, every time.

  • Does not decide eligibility

    It prioritizes candidates and documents the rationale. Eligibility and enrollment decisions belong to the site’s clinical team, in accordance with ICH-GCP.

  • Does not persuade

    It explains options and resolves practical doubts. A decline is accepted without counter-argument. It never promises benefit.

  • Does not improvise under uncertainty

    It is calibrated to prefer “needs review” over “eligible”. The ambiguous case is flagged and resolved by a person.

04 · The backing

Traceable, validated, with the decision where it belongs.

The guarantees are not promises: they are verifiable properties of the system and clauses in the service agreement.

  • R1

    Every output, signed

    Every assessment is recorded with the model version, the fingerprint of the logic, and the rule framework in force at that moment. A period’s evidence is issued as a dossier with its own fingerprint, declaring how much it contains and what it does not guarantee.version · fingerprint · immutable audit trail

  • R2

    Blocking validation

    No version goes into operation without passing validation against reference clinical cases. If it fails, it does not ship. Every coordinator correction feeds the next validation round.gate before every deployment

  • R3

    The decision is human

    The system proposes with a documented rationale; the clinical team decides. That is how it is designed, and how it is written into the contract.ICH-GCP · good clinical practice

  • R4

    Personal data protected

    A dedicated installation per site, in the jurisdiction the site defines. Encryption at rest, explicit consent before any conversation, and revocation that erases the data while preserving the evidence that it was revoked.Law 1581 (CO) · Laws 19.628 and 21.719 (CL) · consent first

Fig. 02 · Coordinator dashboardhuman validation
Chain Reaction coordinator dashboard: a map of barriers detected in conversation and the human validation queue
synthetic data · every decision carries model version + logic fingerprint
05 · The implementation

Implemented and customized. Never an experiment.

The service starts by measuring the site’s real numbers and is configured protocol by protocol. The benchmark is defined together, before operation begins.

  1. Baseline audit

    We measure the site’s actual funnel: where candidates come from, where they drop off, and how long each stage takes. Everything that follows is compared against it.

  2. Per-protocol customization

    Each study’s criteria are loaded and validated with the team. The operation adapts to the site’s real workflow, not the other way around.

  3. Validated operation

    The agents operate, the team validates, and every correction improves the next version, with progress reports against the baseline.

06 · The company

Who is behind it.

Chain Reaction Apps was founded in 2026 by Nicolás Fierro and operates under Chain Reaction Apps LLC, with a single product owner and short cycles: what a site needs becomes product in days, not quarters. Our commercial representation in Colombia works directly with clinics, healthcare institutions, and research sites.

  • FounderNicolás Fierro · 2026
  • EntityChain Reaction Apps LLC (USA)
  • OperationsRemote · Colombia and Chile
  • ModelPer-site implementation + monthly operation
  • DemoOn-site or online, with synthetic data

See the instrument in operation.

contacto@chainreactionapps.com
Book a demo