Why a portfolio-intelligence layer for CAAIN
Every reply, every cross-match, every signal compounds.
Make Data Speak Human
A network funder runs the same operating loop hundreds of times a year: read an inbound, cross-reference the portfolio, surface alignment, decide. Today that loop lives in one CEO's head and a tab of bookmarks. Zeus turns it into shared, durable knowledge that compounds with use.
ALDC is a data processing firm that helps organizations' data speak human. We turn fragmented operational signals — funded projects, member networks, public filings, meeting transcripts — into an institutional knowledge layer that gets smarter every day.
Day 1
The 41 funded CAAIN projects, ISED + AAFC + PrairiesCan filings, and member-network signals are ingested and indexed. Cross-source alignment begins immediately — the same inbound that took an afternoon yesterday takes minutes today.
Day 90
Portfolio patterns emerge. Which projects compound, which regions are under-served, which sectors (fresh produce, food waste) are unfunded gaps the next inbound could fill. Outcome reporting for ISED becomes a rollup, not a scramble.
Year 1
Institutional memory no successor can lose. Every alignment memo, every coordination decision, every funder conversation becomes context for the next one. CAAIN's coordinating function gets sharper the longer the system runs.
Beyond frontier AI
What ChatGPT, Claude, and Gemini can't do for CAAIN's portfolio
Frontier models are great at summarizing documents and drafting prose. They cannot do the four things that matter most for a portfolio-coordinating organization like CAAIN.
No knowledge of CAAIN's 41 projects
They don't know Moove, P&P Optica, F3, MacDon, or the 37 others. Every answer starts from a generic frame — not a portfolio-aware one.
No live operational signal
They can search the web, but they can't connect to project records, ISED filings, the member network at network.caain.ca, or Avoma meeting transcripts. The connective tissue is missing.
No cross-source synthesis
They don't know how an inbound EOI maps to existing funded projects, or how regional gaps connect to sector associations CAAIN already works with. They can't surface the alignment Darrell wrote by hand on May 18.
No institutional memory
Every session starts from scratch. They can't build on what the team learned last quarter, last year, or across the cumulative interactions with funders, partners, and member orgs.
What ALDC adds: a persistent institutional layer that ingests your operational data, learns from every interaction, and grounds every answer in citations to your actual portfolio. The frontier model is still the front-end — but the memory is yours.
AI model routing
The right model for the right job — not one model for everything.
There is no single best AI model. Classification is cheap and should be cheap; user-facing answers should be high-quality; compliance content should be the highest-consequence model. Zeus routes each task to the model that fits.
- Classify an inbound signal (project / news / EOI / question)Claude Haiku
Classification is high-volume + low-stakes. Haiku is fast and ~10× cheaper than the larger models.
- Extract entities (orgs, $, regions, projects) from CAAIN newsSonnet via Batch API
Structured extraction at scale. 50% discount via Batch API; same quality as live.
- Draft an Alignment Memo for the CEOClaude Sonnet 4.6
User-facing prose in the CEO's voice. Quality model with grounded RAG over the portfolio.
- Outcome rollup for ISED quarterly filingClaude Sonnet 4.6 + grounded RAG
Government-facing content. Every claim must be cited; <5% hallucination target enforced.
- Pre-publish review of any portfolio-facing answerClaude Opus 4.7
Highest-consequence gate. Opus catches inversions and over-claims a smaller model misses.
- Cross-portfolio semantic searchVoyage 3 + pgvector
Embeddings are a different job than generation. Specialized model for the retrieval step.
Routing is policy-driven and auditable. Every answer carries a record of which model generated it, which sources it cited, and which review gate it passed.
The stack
Zeus Chat in front. Zeus Memory underneath. Eclipse ingesting the world.
Conversational front-end
The interface CAAIN talks to. Multimodal by default — answers include charts, project cards, source citations, and gap callouts inline, not just text. Routes each question to the model that fits, then grounds the answer in the portfolio.
Context layer for AI
The institutional knowledge layer underneath Chat. Every analysis, every cross-reference, every team interaction is stored and connected. The system gets smarter the longer CAAIN uses it — this is the layer no standalone AI has.
Data Platform
The ingestion + structuring layer that feeds Memory. Pulls in project records, public filings, press releases, member-network activity, and meeting transcripts; normalizes them into a single linked layer. The foundation everything else builds on.
See it on CAAIN's actual portfolio.
Book a 30-minute working session — we'll walk through the Alignment Memo demo using a real inbound (your choice) and a slice of CAAIN's real portfolio. No EOI; this is operational software, not a research project.