Your data. Your pipeline.
Data aggregation and AGI reasoning for any market.

Three pieces. Built for your industry.

Every build has the same shape: a data engine pulling from where the truth lives, a structured vault the AI consults, and an outbound layer that writes off the data in real time. The example below reflects the industry you picked above.

Four panels from a working system.

This is what your operations team would see. Job feeds streaming in, data automation health glanceable, replies landing, the vault graph alive as agents read it.

Recruiting build · live in production.

Below is the in-house tool a recruiting client uses every day. Job database, candidate pipeline, outbound — all running off data pulled directly from the employers' own websites. The page rebuilds when an industry is selected above; this is the receipt that the system ships.

Four phases. Four to six weeks.

Each phase has a deliverable you can read. No vague status updates.

One loop. Four stages.

The product is the loop, not a single slice of it. Aggregate the data, enrich and score it, match with reasoning that shows its work, then act on the result. The match stage is the one most tools skip, so it leads here.

Call it from code. Call it from an agent.

Same engine, same match, two ways in. A REST API for your apps and an MCP server for your AI agents. Agent-native and developer-native at once.

curl https://api.cestoneai.cloud/v1/match \
  -H "Authorization: Bearer $KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "subject": "candidate:ac8f",
    "against": "market:legal-nova",
    "explain": true
  }'

{
  "score": 0.91,
  "verdict": "strong fit",
  "because": [
    "role match: senior litigation",
    "region match: NoVA",
    "2 of 3 hard reqs met"
  ],
  "against": ["1 soft gap flagged"]
}

# 1. agent registers itself (DCR, no shared key)
# 2. OAuth 2.1 + PKCE -> short-lived token
# 3. agent calls a tool
tool: explain_match
args:
  subject: "candidate:ac8f"
  against: "market:legal-nova"

# auth: OAuth 2.1 + PKCE + Dynamic Client Reg (live)
# tools: match · explain_match · score_record
#        get_firm · lateral_moves · list_feeds
# re-point against: any market the engine indexes

{
  "score": 0.91,
  "verdict": "strong fit",
  "because": [
    "role match: senior litigation",
    "region match: NoVA",
    "2 of 3 hard reqs met"
  ],
  "against": ["1 soft gap flagged"]
}

The flagship build, in numbers.

The legal Hiring Index is the one live dataset, and it is the reference build the engine grew up on. It is proof of the engine, not the product: the same loop re-points to any market by swapping the sources. Real metrics, anonymized. No logos, no named customers, design-partner stage.

What we will and will not claim.

The hard part of a data system is trust. Here is exactly where we stand, with the work still in progress labeled as such.

Things every owner asks.

Three ways to start.

Sprint to test the wedge. Build for the full pipeline. Embedded if you want us inside the team.