All work

Finela

Private Capital / Fund Intelligence · Agentic AI

A platform that connects the firm’s own document sources, pulls the figures out and scores its own confidence in each one, keeps every number traceable back to the file it came from, and lets agents analyse and draft against the result.

[ fig. 1 ]A walkthrough of Finela: data sources connected, portfolio questions answered with their sources cited, and a report drafted with charts. Recording, 0:15–0:39.

90 / 100

The confidence an extracted figure needs to pass without a review card. Anything below it is put to a person to confirm.

The problem

Fund data lived wherever it landed — inboxes, shared drives, document stores. Answering an investor question meant hunting through files by hand, and nothing could show where a number had actually come from.

Before
Hunting through files
After
Answers with their sources

How it’s built

Trace one figure backExample figure
Drafted LP report

Revenue for the period was 9,200,000 Q3 pack · C12

Extracted figureRevenue · 9,200,00092 · directRead verbatim from a labelled cell
Low confidenceEBITDA48 · inferred→ asked to a reviewer first
Source cellColumn C, row 12
Source fileQ3 portfolio pack.xlsxSharePointOneDriveGoogle DriveGmailOutlook
[ fig. 2 ]A drawing of the system, not a screen. Swipe sideways to see all of it.

In numbers

190
API endpoints
32
service modules
90
confidence to skip review

Stack

Built with

  • Next.js
  • TypeScript
  • Python
  • FastAPI
  • PostgreSQL
  • Qdrant
  • Redis
  • Azure
  • MCP

Connected to

  • SharePoint
  • OneDrive
  • Google Drive
  • Dropbox
  • Gmail
  • Outlook
  • Slack
  • DealCloud
  • Carta
  • Datasite
NextBiowellHealthcare / Wellness