BUSINESS INTELLIGENCE
Harbinger AI
Harbinger AI is a market intelligence assistant built for commodity and ingredient businesses. It watches thousands of sources, classifies what actually matters for a company, and delivers it as weekly briefings, real-time alerts, and answers to natural language questions. Every claim traces back to a cited source.
Industry
Business Intelligence
Status
Live in production
The challenge.
Commercial teams in food ingredients, chemicals, and agricultural trading spend six to ten hours a week reading news, regulatory filings, and competitor updates by hand. The work is slow, easy to fall behind on, and hard to share across a team. Harbinger needed a way to monitor the same sources automatically, judge which signals were worth acting on, and present them with enough context that a sales or strategy lead could act without double checking everything themselves.
What we built.
We built a conversational assistant on top of a RAG pipeline so every answer is grounded in real source material with dates and links. The backend ingests company data, news, RSS feeds, patent filings, and regulatory databases, then classifies each signal by significance before it reaches a user. Around that we built an admin console for managing tracked companies, signal types, scrape cadence, and the prompts that drive alerts and briefings.
Signals Dashboard
A live read on competitor activity
The dashboard rolls up every signal collected in a chosen period. It shows total signals sent, active companies, and recent activity at a glance, then breaks signals down by company and by type. The recent signals table lists each event with its category, summary, source, date, and a significance rating so the most important items surface first.
- Signals grouped by company and by type with selectable time periods
- Significance ratings flag high priority events for fast triage
- Every signal links back to its original source
Prompt Management
Tunable prompts behind every signal
Admins manage the prompts that decide how signals get classified and how briefings get written. Each prompt covers a category such as patents, product changes, events, or weekly briefings, with its own context and significance guidelines. This keeps classification consistent and lets the team adjust how the AI reasons about each type of news without touching code.
- Separate prompts for alerts, briefings, and each signal category
- Significance guidelines drive consistent severity scoring
- Editable in the console so behaviour can be tuned over time
Company Management
Controlling what gets tracked
The company management screens let admins add and organise the companies being watched, tagged by type and category, with ingestion and active status. A CoreSignal integration finds and imports company data for firms that are not yet ingested, so the tracked set can grow without manual data entry.
- Add, filter, and organise tracked companies and competitors
- CoreSignal integration imports company data on demand
- Ingestion and active status visible per company
Conversational Interface
Ask anything, get a sourced answer
The chat assistant is the front door to the product. A user can ask for a report on top competitors, recent news about a specific company, key industry trends, or recent patent filings, and get a grounded reply. Suggested prompts make the common questions one click away, and the assistant is honest when there is no relevant signal rather than inventing one.
- Natural language queries over competitors, trends, and patents
- Suggested prompts for the most common requests
- Answers stay grounded in retrieved sources, not guesses
How we built it.
The path from first conversation to a production system.
Source mapping
Identified the news, RSS, patent, and regulatory sources that matter for commodity and ingredient markets and built ingestion around them.
RAG pipeline
Built retrieval and classification so each signal is scored by significance and every answer is tied to cited source material.
Admin console
Created the management layer for companies, signal types, scrape cadence, and the prompts that drive alerts and briefings.
Chat and delivery
Wired the conversational interface, dashboard, weekly briefings, and real-time alerts into a single product and shipped it to production.
The results.
What Harbinger AI delivers in production, and the core stack it runs on.
Dashboard, conversational chat, weekly briefings, and real-time alerts in one product.
Every claim and recommendation traces back to a named source with dates and links.
News, RSS feeds, patent filings, regulatory databases, and CoreSignal company data.
Running for commodity and ingredient businesses with an admin console to manage tracked companies and prompts.
Next.js
React
TypeScript
Tailwind
Node.js
PostgreSQL
Vercel
OpenAIHave a project like this?
Tell us what you're building. We'll come back with a clear plan, a timeline, and the team to ship it.
