All projects

Bixso AI domain proof · B2B sourcing & procurement

A product that teaches
B2B sourcing & procurement.

Project bixsourcing Live in production

A B2B sourcing platform where a team of AI agents finds, scores, negotiates and settles deals between buyers and suppliers.

WhyBusiness modelBusiness analysisDomain brainArchitectureThe agents

Why we built it

The reason this project exists.

Sourcing is enormous, manual and slow — buyers and suppliers lose weeks finding each other and getting a deal across the line. It is the ideal proving ground for agentic AI: real rules, real money, clear value. So we built it.

Every BIXSO project starts the same way — with a real business problem worth solving, not a technology looking for a use. Then comes the model, the analysis, and the agents.

01 · Business model

It starts with the business.

Sourcing is a two-sided business: buyers with a need, suppliers with capacity — and value is created the moment the right two meet at the right terms.

  • Buyers raise requests (RFQs); suppliers list catalogues and capacity.
  • Value is created by matching well and fast — and lost to slow, manual search.
  • Free to join; the platform earns on the AI work it does, not on the trade itself.

02 · Business analysis — the DNA

Then the rigour.

Matching well is not a search box — it is a domain with real processes, messy data and hard rules. We modelled the whole deal lifecycle before building anything.

The businessB2B sourcing & procurement
01 Processes
  • Match buyers ↔ suppliers
  • Source & normalise catalogues
  • Negotiate in a live deal room
  • Monitor logistics & ETA
  • Settle via smart escrow
02 Data
  • Specs, units & tolerances
  • Supplier capability & credit
  • Pricing & lead times
  • Compliance & certifications
03 Rules
  • Hard constraints (cert, region, MOQ)
  • Confidence thresholds
  • Escrow release conditions
  • Fraud signals
04 Decisions
  • Which suppliers qualify
  • How to rank trade-offs
  • When to release payment
  • When to ask a human
A modelprecise enough for agents to run

fig. — business analysis: decoding bixsourcing into processes, data, rules & decisions

The rules that let an agent act without going wrong

  • A match shows only above a confidence threshold — below it, the agent asks; it doesn’t guess.
  • Hard constraints (certification, region, minimum order) can never be traded away.
  • Escrow releases only on confirmed logistics milestones — money follows proof, not promises.

02¼ · Domain brain

The product becomes training ground for Bixso AI.

bixsourcing is not only a product page in the portfolio. It is a field-specific operating surface: every workflow, record, decision, exception and correction can become evidence for a specialist domain brain.

BIXSO’s moat is the loop: build real products, learn from real activity, distil the field into an SLM, then place that specialist model inside governed agents. The claim matures only as the product evidence matures.

bixsourcingB2B sourcing & procurement
Domain evidenceprocesses, data, rules, decisions
Distilled SLMsmaller model for one field
Bixso AI braingoverned agents for this domain

02½ · The architecture

The shape of the agentic AI.

Pattern Supervisor–Worker orchestration One master agent plans the deal; seven specialist workers execute it as tools.

Sourcing is a multi-step deal, not a single answer — so it runs on a supervisor. A Sourcing Master Agent classifies intent and role, decomposes the request into a plan, then works a Perceive → Reason → Plan → Act → Verify loop, calling each specialist worker as a typed tool and observing the result before the next step. Crucially, the language model never invents a supplier, a price or a match score: those come from a proprietary Sourcing Knowledge Graph and a deterministic scoring verifier. The LLM reasons and writes; the ground truth is checked.

fig. — supervisor–worker orchestration, the pattern behind bixsourcing

01

Plan → Act → Verify

The master decomposes the request, calls a worker, observes the result, and loops until the goal is met — autonomy with a checkpoint at every step.

02

Workers are tools, not chatbots

Each of the seven workers exposes a typed schema the supervisor calls. New capability is a new tool — testable, auditable, extensible.

03

Grounded, never guessing

Suppliers, prices and % match come from the Knowledge Graph and a deterministic scoring verifier — the LLM never fabricates data.

04

Human-in-the-loop by risk

Read and suggest are autonomous; anything that moves money or leaves the platform requires a human to confirm.

03 · The agents at work

It ships as agentic AI — verified.

The product runs as a team of specialist AI agents — a supervisor routes each step of the deal to the agent built for it.

OrchestrationA Master AI Supervisor orchestrates the deal end to end, handing each stage to the right specialist.

A01agent
Matching agent

Scores buyer–supplier compatibility in real time on industry, region, reputation and price.

A02agent
Sourcing agent

Normalises RFQs, auto-categorises catalogues, forecasts prices and suggests qualified suppliers.

A03agent
Client agent

Scores supplier creditworthiness, reads partner behaviour and flags transaction risk.

A04agent
Deal agent

Drafts contracts, analyses terms, flags legal risk and recommends a negotiation price.

A05agent
Logistics agent

Tracks shipments, detects delays, forecasts ETA and proposes alternative routes.

A06agent
Finance agent

Triggers smart-escrow release on milestones and detects cash-flow anomalies and fraud.

A07agent
Strategy agent

Runs SWOT on each side, forecasts market trends and surfaces cross-sell opportunities.

What the product does

  • Pair buyers and suppliers with an explainable score
  • Run a live, AI-assisted negotiation
  • Hold and release payment by smart escrow
  • Watch logistics and flag risk autonomously

Guardrail. Agents act within guardrails — they match, draft and monitor; a human signs and approves. Every score is explainable; hard rules are never bent.

Live in production

Live at bixsourcing.com — built on Firestore and Google Cloud Run, with a full end-to-end test suite.

Visit bixsourcing →

Frequently asked

Questions about bixsourcing.

What is bixsourcing?

A B2B sourcing platform where a team of AI agents finds, scores, negotiates and settles deals between buyers and suppliers.

Is bixsourcing available now?

Live in production. Live at bixsourcing.com — built on Firestore and Google Cloud Run, with a full end-to-end test suite.

How does bixsourcing use agentic AI?

The product runs as a team of specialist AI agents — a supervisor routes each step of the deal to the agent built for it. A Master AI Supervisor orchestrates the deal end to end, handing each stage to the right specialist.

How does bixsourcing contribute to Bixso AI?

bixsourcing is a field-specific product. Its workflows, records, decisions, exceptions and feedback form domain evidence that can help BIXSO distil a specialist SLM and a governed Bixso AI brain for B2B sourcing & procurement.

What keeps bixsourcing safe and accurate?

Agents act within guardrails — they match, draft and monitor; a human signs and approves. Every score is explainable; hard rules are never bent.

What it means for you

Want this — grounded in your business?

bixsourcing is the BIXSO method, shipped: a business model, analysed with rigour, running as governed agentic AI. We start with the questions, not the tech.

Book a free consult