Home
Ashavid
Back to Portfolio
Built by AshaVidAI / Startup EnablementBuilt 2025

North Road AI

A multi-agent platform that splits founder support into five specialist roles, each with its own knowledge boundary, because a single general assistant gives generically useful answers to specific problems.

Live product
RAGNavigatorBuilderLedgerCounselRainmaker
Five agents, five retrieval boundaries — specialisation is a knowledge scope.

  1. Interface

    What the user touches

    • Next.js 16
    • Tailwind CSS v4

  2. Application

    Where the rules run

    • TypeScript

  3. Intelligence

    Models and ingestion

    • Google Gemini

  4. Data

    What persists

    • Firebase / Firestore

  5. Platform

    What it runs on

    • Stripe

A compact view of the venture

Venture
North Road AI
Category
AI founder copilot
Stage
Live platform
Core challenge
General assistants give general answers
Our role
Product design and full-stack build
Focus areas
Multi-agent architecture, RAG, mentorship, gamification

Why we started building this

Founders ask a wide range of questions, but they ask them from inside a specific role at a specific moment: a pricing question is not a legal question, and neither is answered well by the same context window.

The prevailing pattern was one assistant with one prompt and one undifferentiated knowledge base. It produces answers that are plausible everywhere and authoritative nowhere.

North Road AI was built to test the opposite structure — five specialists with separate remits and separate access to knowledge, so that a financial document informs the finance agent and not the marketing one.

A single assistant that knows everything answers every question at the same depth. Founders do not need average depth — they need the right depth in one area at a time.

What we built

A Next.js 16 platform on Firebase where five specialist agents share a RAG pipeline but not a knowledge scope, wrapped in founder tooling, a mentor portal, and Stripe-ready monetisation.

  • Five specialist agents

    Navigator (strategy and operations), Builder (product and engineering), Ledger (finance), Counsel (legal and compliance), and Rainmaker (sales and marketing).

  • Scoped knowledge access

    Admins assign documents to specific agents, so financial reports reach Ledger without leaking into every other conversation.

  • RAG over mixed sources

    Retrieval across uploaded PDFs and text, scraped websites, and ingested Hugging Face datasets.

  • Startup DNA profile

    Structured tracking of burn rate, runway, stage, and key metrics that agents can reason against.

  • Mentorship portal

    A dedicated mentor dashboard for overseeing and guiding founder progress.

  • Founder Score and leaderboards

    Gamification that gives progress a visible shape across a long and unstructured journey.

Why this needed to exist

  • One assistant with one knowledge base answers narrow questions with broad context, which reads as confident and vague.

  • Not all founder documents should inform all conversations — knowledge scope is a correctness requirement, not a permissions nicety.

  • RAG quality depends almost entirely on ingestion, which is the least glamorous part of the system.

  • Founder progress is long, non-linear, and easy to lose track of without some external structure.

  • Mentors need a different view of the same data than the founders they are guiding.

How we framed the opportunity

The core architectural bet is that specialisation with bounded knowledge beats generality with a larger context window.

Agents are a knowledge-scoping mechanism

The five agents are not personas over one model — they are five retrieval boundaries, which is what makes their answers differ meaningfully.

Ingestion is the real product surface

Document upload, web scraping, and dataset ingestion got first-class admin tooling because retrieval quality is capped by what got indexed.

Structured founder state beats conversational memory

Startup DNA holds burn, runway, and stage as data the agents can read, rather than hoping it surfaces in chat history.

Real-time sync suits an advisory product

Firestore was chosen so mentor and founder views of the same engagement stay consistent without polling.

Where the existing options fell short

What general assistants do well

  • Broad coverage across every topic a founder might raise.
  • Very low friction — no setup, no configuration, no ingestion step.
  • Strong general reasoning when the question is self-contained.

What they cannot do

  • Reason from the founder’s own documents and metrics.
  • Keep sensitive context scoped to the right domain.
  • Give a mentor visibility into the founder’s trajectory.
  • Hold structured state about the company across sessions.

The decisions that shaped the product

Specialise the agents, scope the knowledge

Each agent owns a domain and sees only the documents assigned to it, which is what makes the specialisation real rather than cosmetic.

Invest in ingestion before generation

Uploads, scraping, and dataset ingestion are admin-grade tooling, because retrieval is only as good as the index behind it.

Give the journey a visible shape

Startup DNA, Founder Score, and the mentor portal turn an unstructured multi-year path into something that can be reviewed.

How it came together

  1. Phase 1

    Agent architecture and RAG pipeline

    Define the five agent domains, build retrieval on Firestore, and establish per-agent knowledge scoping.

  2. Phase 2

    Knowledge admin and ingestion

    Ship document upload, the web scraper, Hugging Face dataset ingestion, and the assignment interface.

  3. Phase 3

    Founder tooling and mentorship

    Layer Startup DNA, chat persistence, the mentor dashboard, gamification, and Stripe-based monetisation.

Where the product stands today

  • Five specialist agents with independently scoped knowledge access
  • A RAG pipeline spanning uploads, scraped sites, and Hugging Face datasets
  • An admin knowledge base with per-agent document assignment
  • Startup DNA tracking for burn rate, runway, stage, and key metrics
  • A mentor portal with oversight of founder progress
  • Chat persistence with resumable sessions, plus Founder Score and leaderboards

What building this taught us

  • Specialisation is about retrieval boundaries, not prompt personalities.
  • RAG products live or die on ingestion tooling.
  • Structured state beats conversational memory for anything that matters.
  • Advisory products need a view for the advisor, not only the advisee.
  • Long journeys need visible structure or people lose the thread.

We build the products we would want to operate ourselves.

North Road AI is one of the ventures we design, build, and run in-house at AshaVid.

Talk to usBack to portfolio