AI Sales Agent
for Excess Stock

Dirac Trading Group | Hackathon Brief

The Dirac Thesis

Given

Excess stock is around to stay. It is a massive market. Tech innovations lead to innovations in retail (Walmart, Amazon). Data which could not be serialised and deserialised now can.

When

We start applying the same technical rigour that companies like Jane Street applied to options trading to the much messier world of commodities trading of branded goods.

Then

A trading desk model. A kernel of high quality software + AI tools. Growth is adding another trading desk.

LLMs at Dirac


Earliest adopters of LLMs in the excess stock industry.


We don't build SaaS. We don't build marketplaces.
We buy and sell goods.


LLMs increase operational efficiency: better margins, more business value.

LLMs at Dirac today


1. Product deduplication

2. User interfaces

3. Messy data parsing

4. Customising automated outbound

A Real Problem at Dirac


We have a reliable source of branded sportswear excess stock.


Our buyers prefer over-the-line communication: phone calls, not emails. But we only have 2 team members on sales, and they can only spend part of their day doing cold outbound.


Most of their calling time is idle, waiting for someone to pick up.


AI could be the solution.

Your Task: Experiment


We want you to experiment with what voice agents can do in sales.


Build a proof of concept.

Context provided

Company briefing: who we are, how we sell, MOVs, lead times, tone of voice

Customer info: lead status, source, past orders, contact details

Live products: deal sheet with pricing, MOQs, availability

Brand catalogue: what we stock, categories, key lines

Call transcripts: real sales calls for tone and flow

Potential features

MCP tools for context

Pull live products, customer history, pricing mid-call

Meeting booking

Qualify the lead, book a follow-up

Follow-up emails

Auto-send a personalised email after the call

Live dashboard

Real-time view of calls, outcomes, next steps

Buyer research

Scrape eBay/Amazon to personalise the pitch

Multi-language

Handle calls in the buyer's language

Call summaries

Log what was discussed back to the CRM

Lead scoring

Rate and prioritise leads based on the conversation

...but build whatever you want.

The Tech


Each team gets a company card with £50 on it.


Use it to sign up for any SaaS you need.


You can (but absolutely don't have to) use a pre-provisioned Twilio number.


Apart from that, it's up to you.

Now go build something.