Skip to main content
ClearPathNEPA

Product walkthrough

A guided tour of the working product

Follow the digital-first path from intake to determination as each of its four audiences sees it, then look at the architecture underneath. Every screen below is the real application.

For applicants · project proponents

Submit once, in plain language, then watch it move

A proponent describes the proposed action in everyday language and drops a pin on a map. ClearPath opens the case, runs an AI completeness check so nothing stalls for a missing field, and from that moment the applicant can track the review all the way to a determination. No phone calls, no black box.

The guided intake wizard, step one of five: a review-type selector, a drop zone that reads a project-description PDF to fill the form, a list of sample projects each offering a preview of its own PDF, and the required title and description fields, with a 'What happens next' panel explaining the AI completeness check, CE screening, and human review.

Intake: a guided wizard for a proposed fiber-optic line in an existing road right-of-way.

  1. Five guided steps. Only a title and description are required
  2. Drop a project-description PDF and ClearPath reads it to fill the form
  3. “What happens next” sets expectations: completeness, screening, human review

  • Low-friction intake

    Only a title and description are required. Everything else sharpens the screen but can wait.

  • Completeness help, not gatekeeping

    The AI flags weak or missing information with specific suggestions. It is a drafting aid, never a wall.

  • Track to the determination

    A live case view shows exactly where the review stands and a projected completion date.

The determination screen showing 'Categorical Exclusion granted', the issuing authority and catalog reference, the applicable CE, an AI-drafted determination rationale the reviewer signed, and machine-readable JSON plus an on-demand PDF.

The end state an applicant tracks to: a structured Categorical Exclusion determination.

  1. The structured determination: “Categorical Exclusion granted”
  2. An AI-drafted rationale, edited and signed by the reviewer
  3. JSON plus an on-demand PDF, generated from one machine-readable record

For agency reviewers

The reviewer retains complete authority

ClearPath is human-in-the-loop by construction. The AI screens the action against all 2,105 federal Categorical Exclusions and proposes matches with a confidence score and the real legal citation. Every output is labeled a recommendation, pending review. The reviewer approves, edits, or overrides each one, with a documented rationale for anything they change, then issues the determination and the administrative record.

The review workbench on one surface: a 'Human-in-the-loop' badge, an AI completeness check, ranked Categorical Exclusion matches with confidence scores and real citations that double as the reviewer's radio choices, per-circumstance approve/override controls, an extraordinary-circumstances / GIS screen, similar past actions, and an AI-drafted determination rationale the reviewer edits before signing.

Screening and the reviewer decision on one working surface: the AI recommends, the human decides.

  1. Human-in-the-loop: every AI output is a recommendation the reviewer approves or overrides
  2. Ranked CE matches with a confidence score and the real legal citation, each one selectable
  3. A lightweight extraordinary-circumstances / GIS screen flags issues before they surface late
  4. ClearPath drafts the determination rationale; the reviewer edits and signs it

  • Authority never leaves the human

    The AI advises; the decision payload records exactly what the reviewer determined.

  • Real citations, not guesses

    Matches are drawn from the actual federal CE catalog, with the legal structure and a source link.

  • Defensible by default

    Issue a determination and a complete administrative record of every input, suggestion, and action.

For agency leadership

Portfolio status, projected completion, days saved

Leadership sees the whole pipeline at a glance, read entirely from the append-only event store. The numbers are the record, not a separate spreadsheet. Each case carries a projected completion date and the days saved against a stated manual baseline; the reviews at risk of slipping are flagged and sorted to the top.

The leadership portfolio dashboard: KPI cards for total, in-review, completed, at-risk, average days to determine, and average days saved; a stage-distribution bar; a case-flow chart plotting every review from intake against a 45-day manual baseline; phase-duration bullet bars; and a case table spanning agencies with the at-risk row highlighted and sorted first.

The leadership portfolio: live status across every CE review, with each case plotted against the 45-day manual baseline.

  1. Days saved vs. a stated manual baseline, quantified on screen
  2. At-risk reviews are flagged and sorted first
  3. A projected completion date for every case, from a per-phase SLA model

  • Projected timelines

    An explicit SLA model plus empirical history gives every case a defensible completion date.

  • Bottlenecks surfaced

    Stage distribution and at-risk highlighting show where the queue is slowing, before it slips.

  • Days saved, in the open

    A transparent comparison against an agency-defined baseline, with the assumption stated, not hidden.

For the public · stakeholders

Consistent records, documented rationale, open data

The fourth audience is the public. Comments and engagements are first-class entities, logged to the same machine-readable record and compiled for reviewer attention. Because the record is structured and provenance-stamped at the moment of decision, it is FOIA- and litigation-ready, and the same case is retrievable as standard JSON through a documented API.

The public participation panel: an AI comment analysis summarizing support, opposition, and concern counts with theme tags across the docket, the individual public comments logged with dates and channels, and the engagement log beneath.

Public participation: comments compiled and logged to the record.

  1. AI compiles public comments into support, opposition, and concerns (functional requirement #8)
  2. Each comment is logged to the same machine-readable administrative record
The engagement log: a recorded public meeting at a Ranger District office with attendance and notes, plus a form to log additional meetings, hearings, and consultations.

Meetings, hearings, and consultations recorded as standard engagement entities.

  1. Public meetings and hearings recorded as standard engagement entities
  2. Log consultations into the auditable public-participation record
The same case retrieved as standard JSON from the data-standard API, showing snake_case fields and the six provenance fields (record owner agency, data source agency, data source system, data record version, last updated, and retrieved timestamp) on each entity.

The same record, retrievable as data-standard JSON with provenance on every entity.

  1. Six provenance fields on every entity: owner, source system, version, timestamps
  2. The same case, retrievable as standard JSON via the documented OpenAPI surface

  • A real place to be heard

    Comments and engagements are captured as standard entities, not lost in an inbox.

  • FOIA- and litigation-ready

    A complete, machine-readable record exists at the time of decision, not reconstructed later.

  • Auditable and open

    Documented rationale and provenance on every record make the workflow inspectable end to end.

Under the hood · the architecture

API-first and event-driven, interoperable by construction

ClearPath is built on the federal NEPA & Permitting Data Standard as its canonical model, with an append-only event store at the core and an OpenAPI surface as the contract. Every AI feature degrades to a deterministic fallback, so the system never hard-fails. The production target is a region-bound GovCloud deployment on the FedRAMP / NIST 800-53 path, with Login.gov identity and Section 508 / WCAG 2.1 AA throughout.

Architecture diagram: an API-first, event-driven system with a Next.js front end and BFF, isolated domain services, an append-only event store over PostgreSQL, AI with a deterministic fallback, and an OpenAPI surface, targeting GovCloud with FedRAMP and NIST 800-53.

The API-first, event-driven architecture: Next.js front end, isolated domain services, an append-only event store, and an OpenAPI contract.

  • API-first, event-driven

    Every state change writes a case event; the OpenAPI surface is the contract, not an afterthought.

  • Data-standard-native

    Prisma entities mirror the 13 standard entities, with the six provenance fields on every record.

  • AI + deterministic fallback

    If the model is unavailable, a deterministic result keeps the interface identical, so the review never stalls.

  • Interoperability by construction

    Partner systems read the same case as standard JSON. Portability is proven, not promised.

  • GovCloud / FedRAMP path

    The production target is region-bound GovCloud on the NIST 800-53 path, Login.gov, and 508/AA.

  • Scales without the edge

    SSG public pages + SSR app on horizontal regional containers, Postgres replicas, and background workers. No edge required.

Data-flow diagram of the seven-step golden path (intake, AI screening, reviewer decision, determination, administrative record, leadership dashboard, and data-standard API read) flowing over the append-only event store.

The seven-step golden path, flowing over the append-only event store.

Data-model diagram of the 13 NEPA & Permitting Data Standard entities and their relationships, each carrying the six provenance fields.

The canonical model: 13 standard entities, every one provenance-stamped.

Before-and-after diagram contrasting a roughly 45-day manual Categorical Exclusion process with ClearPath's minutes-long digital-first path, showing an estimated ~80% reduction in administrative effort, modeled against the ~45-day manual baseline.

The impact: a weeks-long manual memo becomes a digital-first determination in minutes, with an estimated ~80% less administrative effort (modeled against a ~45-day manual baseline).

Walk it yourself

Everything above is the live application. Open it, switch roles in the masthead, and run the same path end to end, or step through the seven-step review first.

Open the live demoSee the seven steps