Skip to content
HMD Digital
Process example · Consultants and ops teams who screen public incentive programmes for SMEs

Autobandi

An example of a process we take on. An Italy-domain process example under the same pattern as the other agents: give the matching desk a company tax ID and a bit of context. It asks clarifying questions, enriches the company when keys allow, then matches against a seeded open-programme corpus (shaped like a national incentive portal; not a live feed today) with deterministic gates, vector shortlisting, and a reasoning filter. You review the shortlist.

Start a discovery conversationAll examples

Pattern we have built

Your process is the product

The problem

Open incentives are public. Finding the ones that fit is not.

Public portals list hundreds of programmes. Eligibility depends on geography, size, sector codes, dates, and typology. Spreadsheets and generic chat tools either miss gates or invent fit. The demo corpus follows an Italy incentive-portal shape; the process pattern is what we take on elsewhere.

Before the agent

Too many open programmes to scan by hand
Hard eligibility gates buried in metadata
Generic AI that invents certainty
No clear rationale to hand a client
What the agent does

What Autobandi handles.

Start from a company ID

Conversational intake with optional clarifying questions (geography, size, sector) before matching.

Enrich, then match

Registry/web context when available, then open programmes from a seeded corpus shaped like a national incentive portal (live ingest is not what the demo loop uses today).

Three filters, one shortlist

Deterministic qualify, intent/vector shortlist, then LLM reasoning with fitness and rationale.

You keep the judgment

Ranked matches with reasons; not autopilot applications.

LangGraph agent loop

The matching desk streams steps so you see clarify, enrich, qualify, shortlist, and reason.

Hard gates, not soft guesses

Built around tax ID, region, size, and sector codes in the demo domain; not a generic grant chatbot.

Where people stay in the loop

The agent does the work. People keep the judgment.

The agent does the eligibility grind. You decide which programmes are worth pursuing.

  1. Company ID inTax ID plus optional open-text context
  2. PersonClarifyAgent asks only what it still needs
  3. EnrichRegistry / web context when keys allow
  4. MatchDeterministic, vector, then reason
  5. PersonYou reviewShortlist with rationale
Agent step A person decides
How it's managed

Watched, improved, accountable.

When we manage an agent for you, we watch performance, costs and exceptions, with enough control to fix what needs fixing.

Corpus source labelledThe demo loop matches a seeded corpus shaped like national incentive-portal records, not a live feed.
Honest soft-skipsIf a registry or LLM key is missing, the loop continues with clear stubs; no invented certainty.
EU-minded processingAnalytics and models follow the same consent and EU host patterns as the other HMD agents.

Stack, briefly

Matching desk with an agent loop over a seeded open-programme corpus.

  • Next.js
  • Supabase
  • LangGraph

Related process for your business

Eligibility work that eats consultant hours?

Same pattern wherever public programmes have hard gates and soft fit.

  • Regional calls
  • Innovation vouchers
  • Training funds
  • Green transition schemes
Tell us about yours

More process examples

MidasCVClient-ready packs from messy CVs and hiring profiles.UltravisibleTurn expertise into a steady, on-brand publishing rhythm. LinkedIn first.DocableDichiarazioni and tender annex packs filled from bound sources.

Got a process you'd happily never think about again?

Tell us about it. One conversation is enough to know whether there's a real opportunity.

Start a discovery conversation