map

Partnership intelligence for university technology transfer. Type a sector, get a fully sourced report that maps public companies to overlapping research at UNC Chapel Hill, scored and citation checked, in about a minute.

Disclaimer. This is an independent project. It is not created by, affiliated with, or endorsed by UNC Chapel Hill or any of its offices. UNC appears only as the analytical subject of the reports the tool generates.


Web app	https://aria-pi-frontend.vercel.app
API	https://aria-pi-api.vercel.app
Status	Live · free tier · no third-party API keys required

What it does

Given a sector (for example Oncology, Semiconductors, or a free-text term like Energy Minerals), map assembles a partnership intelligence report:

Resolves the sector to a company set (curated top companies plus NC-based firms, or live SEC EDGAR full-text discovery for niche terms).
Pulls primary-source data per company, in parallel, from SEC EDGAR, ClinicalTrials.gov, PubMed, and NIH RePORTER.
Builds a deterministic, 7-section report plus a one-page executive summary, with every claim backed by at least two citable URLs.
Streams real progress to the browser as each company resolves, then renders an interactive report with charts and a live table of contents.
Opens the same data in a multi-view workspace: Report, Visualize (23 charts led by a rotating 3D connection orbit, plus network/Sankey diagrams), Trends (10-year SEC financial trajectories), Excel (18-sheet analytics workbook), and Slide Deck (speaker-noted, per-sector deck).
Exports to Markdown, PDF, Word, Excel (.xlsx), and PowerPoint (.pptx), with the on-screen visuals preserved.

Every factual claim carries its source URLs. Anything that cannot be double-sourced is flagged for analyst review rather than guessed.

Key features

Live progress streaming. The backend emits Server-Sent Events as each company completes, so the progress UI reflects genuine work, not a timer.
Sector resolution. 24 curated sectors (top 15 global + NC-based seeds) plus live SEC EDGAR discovery for any other term, with a clear resolution label (curated / discovered / default).
Source-verified claims. Two-source rule, domain blocklist (no Wikipedia / aggregators), and a per-report verification count.
One-page executive summary. Metric tiles, thesis, pie charts, a financial snapshot, NC context, and active UNC units, first in the report.
Interactive report. Inline AMA citations, data visualizations distributed through the sections, and a floating scroll-spy table of contents.
Multi-view workspace. A top nav (Home / Report / Visualize / Trends / Excel / Slide Deck) turns one report's data into five sector-customized views that work like a normal website.
Visualize. 23 unique, per-sector charts led by a rotating 3D connection orbit (companies circling a central UNC node, logo-style, hover to pause), plus a 3D isometric scatter, a connection network, a Sankey flow, correlation matrix, Lorenz curve, Pareto, box plots, and a radar. Soft gradient fills, rounded bars, and gentle card shadows give the page a welcoming feel.
Trends. Stock-style 10-year SEC financial trajectories (revenue, R&D, net income) showing what grew and what shrank, with CAGR and momentum, and thin-coverage years dropped so the latest partial fiscal year never distorts.
Excel workspace. An 18-sheet analytics workbook (HHI concentration, correlation matrix, quartiles, CAGR, partnership-priority scores, segments) with on-screen analytics and live clickable worksheet previews.
Slide deck. A bullet-driven, graduate-level, per-sector deck with speaker notes, exportable to PowerPoint.
High-fidelity exports. Markdown (editable, linked), plus PDF, Word, Excel, and PowerPoint that capture the rendered report so they look like the web page.
Responsive. Lays out cleanly on phones and tablets (iPhone / iPad): the nav wraps, tables scroll horizontally, and SVG charts scale to width.
Animated intro. A network-graph splash that builds out from a central node before handing off to the app.

Architecture

flowchart LR
    U[Browser] -->|POST sector| NX[Next.js route handlers]
    NX -->|proxy SSE / JSON| API[FastAPI orchestrator]
    API -->|parallel fetch| SEC[SEC EDGAR]
    API --> CT[ClinicalTrials.gov]
    API --> PM[PubMed E-utilities]
    API --> NIH[NIH RePORTER]
    API -->|deterministic build| RB[ReportBuilder]
    RB -->|JSON report| NX
    NX -->|render| RPT[Report component]
    RPT -->|views| VW[Report / Visualize / Trends / Excel / Slides]
    VW -->|html2canvas + xlsx + pptxgenjs| EX[PDF / Word / Markdown / Excel / PowerPoint]

Two independently deployed Vercel projects:

aria-pi-frontend - Next.js app. Client renders the report; server-side route handlers (/api/run-pipeline, /api/run-pipeline-stream) proxy to the backend so the browser never calls it directly and no stale cache is served.
aria-pi-api - FastAPI app on the @vercel/python runtime. Resolves the company set, fans out to the four public data APIs concurrently within a hard time budget, and assembles the report deterministically.

Request lifecycle (streaming path):

sector ─▶ /api/run-pipeline-stream ─▶ /run-pipeline-stream
        ◀─ stage:resolved (total)
        ◀─ progress (1/N company)  ... per company as it completes
        ◀─ stage:building ─▶ stage:verifying
        ◀─ done (full report JSON)

If streaming is unavailable, the frontend falls back to the plain /run-pipeline request plus a cosmetic progress animation, so a report always loads.

Tech stack

Layer	Technology
Frontend	Next.js 14 (App Router), React 18, TypeScript 5
Charts	Hand-rolled inline SVG (no chart library), incl. animated 3D
Exports	`docx`, `jspdf`, `html2canvas`, `xlsx` (SheetJS), `pptxgenjs`
Backend	FastAPI, Pydantic, `requests` (Python 3.12)
Runtime	Vercel (Next.js + `@vercel/python` serverless)
Data	SEC EDGAR, ClinicalTrials.gov, PubMed, NIH RePORTER
Optional	Anthropic Claude (legacy synthesis path; off by default)

Repository layout

research-graph/
├── aria-pi-frontend/                  # Next.js web app  (Vercel: aria-pi-frontend)
│   ├── public/
│   │   └── flow.html                  # standalone system data-flow diagram
│   └── src/
│       ├── app/
│       │   ├── page.tsx               # intro + search + streaming progress + report mount
│       │   ├── layout.tsx
│       │   ├── globals.css
│       │   └── api/
│       │       ├── run-pipeline/route.ts         # JSON proxy to backend
│       │       └── run-pipeline-stream/route.ts  # SSE proxy to backend
│       ├── components/
│       │   ├── Report.tsx             # report renderer, charts, TOC, summary
│       │   ├── VisualsView.tsx        # Visualize page: 23 charts + diagrams
│       │   ├── TrendsView.tsx         # Trends page: 10-year SEC trajectories
│       │   ├── Chart3D.tsx            # rotating orbit + isometric 3D scatter (SVG)
│       │   └── Intro.tsx              # animated network-graph splash
│       └── lib/
│           ├── report-analytics.ts    # HHI, correlation, quartiles, scores
│           ├── report-excel.ts        # 18-sheet .xlsx workbook builder
│           ├── report-slides.ts       # per-sector .pptx deck + speaker notes
│           └── report-export.ts       # Markdown / PDF / Word exporters
│
├── aria-pi-backend/                   # FastAPI service  (Vercel: aria-pi-api)
│   ├── api/index.py                   # Vercel ASGI entry point
│   ├── vercel.json                    # @vercel/python build config
│   └── aria_pi/
│       ├── orchestrator.py            # FastAPI app + endpoints + concurrency
│       ├── sectors.py                 # sector resolution, curated + NC seeds
│       ├── clients/
│       │   ├── sec_edgar_client.py    # facts, XBRL financials, discovery, DEF 14A
│       │   ├── clinicaltrials_client.py
│       │   ├── pubmed_client.py
│       │   ├── nih_reporter_client.py
│       │   └── claude_client.py       # optional LLM path (off unless key set)
│       ├── builders/report_builder.py # deterministic 7-section assembly
│       ├── utils/source_tagger.py     # 2-source validation + blocklist
│       ├── models/                    # Pydantic models
│       └── data/                      # curated UNC units, datasets, programs
│
└── README.md

Note: top-level src/ and aria_pi/ directories are a stale early prototype and are gitignored. The shipped apps live under aria-pi-frontend/ and aria-pi-backend/.

Data sources

All free, all primary-source, no API keys required.

Source	Provides	Endpoint
SEC EDGAR	Company facts, XBRL financials, filings, full-text discovery, DEF 14A proxies	`data.sec.gov`, `efts.sec.gov`
ClinicalTrials.gov v2	Sponsor-matched trials, phases, collaborators	`clinicaltrials.gov/api/v2/studies`
PubMed (E-utilities)	UNC co-authored publications by school	`eutils.ncbi.nlm.nih.gov`
NIH RePORTER	Active grants mentioning a company + UNC	`api.reporter.nih.gov/v2/projects/search`

API reference

Base URL: https://aria-pi-api.vercel.app

`GET /status`

Health and mode info.

`POST /run-pipeline`

Build a full report and return it as one JSON payload.

// request
{ "sector": "Oncology", "companies": ["Merck", "Pfizer"] }  // companies optional

// response
{ "sector": "Oncology", "status": "COMPLETED", "data": { /* full report */ } }

`POST /run-pipeline-stream`

Same pipeline, streamed as Server-Sent Events (text/event-stream). Frame types:

`type`	Payload	Meaning
`stage`	`{ key: "resolved", total, resolution }`	company set resolved
`progress`	`{ done, total, company }`	one company finished fetching
`stage`	`{ key: "building" }`	assembling the report
`stage`	`{ key: "verifying" }`	source validation
`done`	`{ report }`	finished report object
`error`	`{ message }`	failure (client falls back)

The report object carries report_meta, section1_overview … section7, references, _validation (claim counts), and _meta (resolution, seeds).

Report structure

Each report contains a one-page Summary followed by seven sections:

#	Section	Highlights
Summary	Executive brief	metric tiles, thesis, pie charts, SEC snapshot, NC context, UNC units
1	Sector Overview	definition, scale, why now, NC context, UNC units; revenue & R&D charts
2	Internal Mapping	known partnerships, faculty, data assets, risk flags; alignment chart
3	Company Selection	selected vs excluded; UNC-tie and partnership-scale pie charts
4	Company Profiles	per-company facts, filings, pipeline, partnering, UNC alignment, signals, UNC alumni
5	Value Proposition	data assets, research capacity, talent, NC access, models
6	Talking Points	sourced per-company outreach points
7	References	AMA-style, de-duplicated, numbered

Workspace views

The top nav exposes five views over the same generated report. Each is customized to the resolved sector and downloadable.

View	What it shows	Export
Report	The sourced 7-section report + one-page summary	Markdown, PDF, Word
Visualize	23 sector-specific charts: rotating 3D connection orbit, 3D isometric scatter, connection network, Sankey, correlation matrix, Lorenz curve, Pareto, box plots, radar, heatmap	(image capture via PDF/Word)
Trends	10-year SEC financial trajectories (revenue, R&D, net income), CAGR, growth momentum; thin-coverage years dropped	-
Excel	18-sheet analytics workbook with live, clickable worksheet previews and on-screen analytics	`.xlsx`
Slide Deck	Bullet-driven, graduate-level, per-sector deck with speaker notes	`.pptx`

Analytics behind Visualize/Trends/Excel (report-analytics.ts): HHI concentration, Pearson correlation matrix, quartile and five-number summaries, CAGR, a 0-100 partnership-priority score, percentile ranks, Lorenz curve, and segment analysis.

Local development

Two services run side by side. Use two terminals.

Backend (FastAPI)

cd aria-pi-backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
uvicorn aria_pi.orchestrator:app --reload --port 8000
# http://localhost:8000/status

Frontend (Next.js)

cd aria-pi-frontend
npm install
# point the proxy at the local backend
echo 'BACKEND_API_URL=http://localhost:8000' > .env.local
npm run dev
# http://localhost:3000

Tests

cd aria-pi-backend && pytest

Environment variables

Frontend (server-side only; never exposed to the browser):

Variable	Required	Description
`BACKEND_API_URL`	No	Backend base URL. Defaults to the live API alias.
`VERCEL_AUTOMATION_BYPASS_SECRET`	No	Set only if the backend project has Deployment Protection on.

Backend:

Variable	Required	Description
`ANTHROPIC_API_KEY`	No	Enables the optional Claude synthesis path. Omitted = deterministic builder (default).
`ANTHROPIC_MODEL`	No	Overrides the default Claude model when the key is set.

Deployment

Both projects deploy to Vercel from the Vercel CLI (GitHub is the source of truth for code; pushes do not auto-deploy these two projects).

# frontend
cd aria-pi-frontend && npx vercel --prod

# backend
cd aria-pi-backend && npx vercel --prod

Frontend: standard Next.js build (aria-pi-frontend/vercel.json).
Backend: @vercel/python builds api/index.py, bundling aria_pi/**, with a 60s function maxDuration (aria-pi-backend/vercel.json).

Performance and limits

Concurrency budget. Up to 22 companies are fetched in parallel under a hard ~44s data-collection deadline so the function stays within Vercel's 60s cap. Companies that miss the deadline keep an SEC-only stub; the report still renders.
Streaming cadence. Progress events fire as each company resolves, giving granular "N of M analyzed" feedback.
Export size. PDF/Word capture the rendered DOM as paginated page images, so large reports (20+ companies) can run ~70-95 pages and take ~20s to build. Markdown stays lightweight, fully text, and linked; Excel and PowerPoint are generated from data and stay small.
Mobile. The UI is responsive on iPhone and iPad. The image-based PDF/Word capture is memory-heavy, so on a phone, large (90+ page) captures are slow and best done on a laptop; Markdown, Excel, and PowerPoint exports are light and fine on mobile. The floating table of contents is hidden below 1380px.

Data integrity rules

Two-source rule. Every claim needs at least two independent citable URLs or it is flagged for analyst review.
Source blocklist. Wikipedia, aggregators, and unattributed news are rejected; SEC, ClinicalTrials.gov, PubMed, NIH RePORTER, and peer-reviewed journals are accepted.
Sponsor matching. Clinical trials are matched on sponsor/collaborator fields, not free-text, so unrelated trials are not attributed to a company.
Clean typography. Generated reports never contain em or en dashes.

Limitations

Private companies (no SEC filings) render partial profiles sourced only from trials, publications, and grants.
UNC alumni detection reads DEF 14A proxy statements and public leadership pages; it covers board members and named executives, not every employee, and LinkedIn cannot be scraped (links are pre-filled searches).
The tool removes mechanical labor; it does not replace analyst judgment. Reports are drafts for human verification before any outreach.

Roadmap

Internal relationship database to replace flat-file UNC seeds.
Report versioning and diffs across time for the same sector.
Optional selectable-text PDF mode alongside the pixel-perfect capture.
CRM-ready structured export.

License

MIT. See LICENSE.

Name		Name	Last commit message	Last commit date
Latest commit History 79 Commits
.claude		.claude
aria-pi-backend		aria-pi-backend
aria-pi-frontend		aria-pi-frontend
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

map

Table of contents

What it does

Key features

Architecture

Tech stack

Repository layout

Data sources

API reference

`GET /status`

`POST /run-pipeline`

`POST /run-pipeline-stream`

Report structure

Workspace views

Local development

Backend (FastAPI)

Frontend (Next.js)

Tests

Environment variables

Deployment

Performance and limits

Data integrity rules

Limitations

Roadmap

License

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

map

Table of contents

What it does

Key features

Architecture

Tech stack

Repository layout

Data sources

API reference

GET /status

POST /run-pipeline

POST /run-pipeline-stream

Report structure

Workspace views

Local development

Backend (FastAPI)

Frontend (Next.js)

Tests

Environment variables

Deployment

Performance and limits

Data integrity rules

Limitations

Roadmap

License

About

Topics

Resources

License

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

`GET /status`

`POST /run-pipeline`

`POST /run-pipeline-stream`

Packages