Mímir 4.0
Viska Agent Department

6-Month Roadmap: From Research Bot to AI Intelligence Infrastructure

For  Guðlaugur Steinarr Gíslason, CIO
From  boas.dev
Engagement  Full-time, 6 months

From chatbot to intelligence department

Mímir has proven one thing: AI-assisted research works for your team. Daði said it's like having two full-time analysts. But Mímir today is one tool solving one problem — broker research search. Your operation has ten problems.

I'm not offering to maintain a chatbot. I'm offering to build a complete AI intelligence department for Viska over 6 months, full-time. At the end of the engagement, you own the system — GitHub repo, all infrastructure, all agents, full documentation. Kristján can contribute code. Any developer can maintain it.

Why now

AI went from differentiator to baseline expectation in 2025 — 95% of fund managers now allow employees to use multiple AI applications. But the tools available today don't fit a 7-person fund:

Option Cost What You Get
Hebbia $10K/seat/yr ($70K team) Generic document synthesis — same as every other fund
AlphaSense $10–20K/seat/yr ($100K+ team) Search over their corpus — not yours
Build in-house $500K–1M+/yr 6–12 months before any value, talent hard to find in Iceland
Do nothing $0 Watch Minotaur Capital (zero analysts, 27% returns FY25) eat your lunch

None of these build what you actually need: a bespoke system tuned to your two funds, your research process, your investment thesis.

What you already have

This isn't starting from zero. This is activating a system that's 60% built.

System Status What It Does
Mímir Chat Live RAG search over 30K chunks from 29 brokers. Thread memory, broker filtering, source attribution
Daily Brief Live Cross-broker synthesis every morning — thematic buckets, high-impact flags, portfolio outlook
Report Generation Live Conversation → polished deliverable. IC-ready, LinkedIn-ready, archive-ready
Viska Portal Live Bloomberg-style 8-section intelligence platform. Chat is mature; 6 more sections coming online
Corpus Pipeline 88% Dropbox → OCR → embed → Supabase. v4.0 fixes shipping now
Trading War Room 70% 4 agent repos, 7/8 n8n automations live, Alpaca + FRED + QuantOracle integrated, risk framework approved
Knowledge Graph Production Wiki engine, bi-temporal graph (22K nodes), decision capture, batch ingest — ~70% reusable for client deployment

6-Month Delivery Plan

Month 1 — Intelligence Foundation

Make the existing system bulletproof and add the interfaces that multiply access.

Deliverable CIO Priority Detail
Ingestion pipeline v4.0 Foundation Fix 3 compounding bugs, verify with full re-ingest. Corpus goes from 2 dates to full history.
Portal data activation Foundation Unblock credential architecture — all 8 portal sections go live with real data
Mímir MCP #1 Research Pipeline Query the entire corpus from Claude Desktop, Cursor, or any AI workspace. Not locked to Slack.
Macro research pipeline #1 Research Pipeline Central bank minutes, earnings transcripts, policy documents into corpus. Claude Opus 4.6 synthesizes cross-asset macro narratives.
Ingestion notifications Foundation Instant Slack alerts when new reports arrive and are indexed

What Viska gets at Month 1: A corpus that covers all your sources (not just broker PDFs), accessible from every tool you use — Slack, portal, Claude Desktop, Cursor. The daily brief pulls from a much richer dataset.

Month 2 — Multi-Agent Analysis + Knowledge Base

Go from one bot answering questions to specialized agents running surveillance — backed by institutional memory.

Deliverable CIO Priority Detail
Viska Knowledge Base Foundation All fund documents ingested into a structured wiki with decision trails, entity relationships, and temporal reasoning. A knowledge graph — not more vector search.
Crypto analysis agent #2 Multi-Agent On-chain metrics, exchange flows, funding rates, liquidation data. Runs autonomously, reports anomalies.
Macro sentiment agent #2 Multi-Agent Cross-asset macro radar — FX, rates, commodities, equity indices. Daily conviction scoring.
Geopolitical risk scanner #2 Multi-Agent Regulatory filings, sanctions, elections, policy shifts. Flags portfolio-relevant risk events.
Earnings call analysis #1 Research Pipeline Structured intelligence from transcripts — key metrics, guidance changes, management tone shifts.
Document comparison #1 Research Pipeline Track price target changes and narrative evolution. "How has Goldman's oil view changed in 3 months?"

What Viska gets at Month 2: Three autonomous agents running 24/7 surveillance on crypto, macro, and geopolitical themes — under €200/month in compute. Plus a knowledge base that goes beyond search.

The knowledge base difference: Mímir today does vector search — find text similar to your question. The knowledge base adds temporal reasoning ("what did we decide before the rebalancing?"), relationship traversal ("all LPs who attended meetings about Fund X"), and decision archaeology ("what was rejected, and why?"). This is the difference between a search engine and institutional memory.

Month 3 — Trading War Room

Move from research to action. Live market intelligence and trade discipline.

Head start: The Trading War Room has 4 sessions of build work behind it — 54 planning documents, 4 agent repos with MCP integrations (Alpaca, FRED, QuantOracle with 63 quant tools), 7/8 n8n automations live, risk framework approved, dashboard designed. Month 3 is deployment and activation, not greenfield.

Deliverable CIO Priority Detail
React dashboard Trading 5-page Bloomberg-style trading interface — portfolio overview, trade journal, market page, session log, risk monitor.
Live market data pipeline #3 Sentiment Tracking Already built in n8n — activate R2 writes, connect to dashboard. Real-time price feeds, volume, volatility.
Grok narrative tracking #3 Sentiment Tracking Real-time X/social sentiment via Grok — catches narrative shifts before they move price.
Technical analysis tools Trading QuantOracle MCP (63 deterministic quant tools, 1,000 free/day). Chart pattern detection, momentum, support/resistance.
Trade journal + risk enforcement Trading Decision chain UI, thesis blocks. Risk framework enforces position sizing, consecutive-loss gates, mandatory reflection.
Paper trading launch Trading Alpaca paper account provisioned. 90-day documented paper period — hard gate before any live capital.

What Viska gets at Month 3: A command center for execution with mechanical discipline. The war room doesn't just show data — it enforces rules.

Month 4 — Communications & Content

Turn the intelligence system outward — LP reporting, thought leadership, brand building.

Deliverable CIO Priority Detail
Investor letter automation #4 Communications Monthly LP letters drafted from portfolio data + market context. Performance attribution auto-generated.
Meeting briefing system #10 Meeting Prep Before every family office meeting: deep research on the prospect. After: personalized follow-up.
Thought leadership pipeline #8 Content 2–3 polished pieces per week — newsletters, LinkedIn posts, short-form analysis. Icelandic + English.
CRM knowledge layer #8 Content LP relationships, meeting history, and follow-ups as entities in the knowledge graph. Pipeline from content → meetings → LP relationships.

What Viska gets at Month 4: The bottleneck isn't performance — it's communication bandwidth. The system handles the writing, formatting, and follow-up. You spend your time in meetings instead of at a keyboard.

Month 5 — Advanced Intelligence

Unique capabilities that no off-the-shelf tool provides.

Deliverable CIO Priority Detail
Scenario modeling #6 Scenario Modeling "What happens to crypto allocations in a stagflation environment?" — Claude runs deep analysis overnight, presents scenarios with evidence.
DD on distribution partners #7 Due Diligence Continuous monitoring of partner prospects — regulatory filings, AUM shifts, personnel changes, news.
Personalized watchlists User experience Per-analyst filtered digests and alerts based on individual coverage areas
Coverage gap detection Research quality System identifies what brokers you're NOT seeing and what topics are underrepresented

What Viska gets at Month 5: The system starts thinking ahead — modeling scenarios, watching partners, identifying blind spots. This is where AI infrastructure starts creating genuine information advantage.

Month 6 — Autonomy & Handoff

The system runs without me. You own everything.

Deliverable CIO Priority Detail
Proprietary signal generation #9 Signal Gen Fine-tuned model on your macro framework. Quantitative signals from qualitative conviction. IP protected.
Trade execution pathway Trading Paper trading → live execution via Alpaca. Human approval required for every trade.
Viska GitHub Handoff Complete codebase — all agents, skills, infrastructure, documentation. Kristján has contributor access.
Full documentation Handoff Architecture docs, runbooks, configuration guides. Any developer can maintain and extend the system.
System health dashboard Operations Real-time status of all agents, pipelines, data flows. Know instantly if something breaks.

What Viska gets at Month 6: A fully autonomous AI intelligence department that you own. GitHub repo, all infrastructure, all documentation. 100% handoff — bring in any developer to extend it, or continue on a maintenance contract.

What this replaces

Traditional Cost What Mímir 4.0 Replaces
2 research analysts (ISK 1.2–1.6M/mo) Automated research pipeline + daily briefs + agent surveillance
Bloomberg Terminal (5 seats × $2K/mo = ISK 1.4M/mo) Portal + real-time data + natural language search
Part-time content writer (ISK 300–500K/mo) Thought leadership pipeline
Meeting prep time (10+ hours/week across team) Automated briefing + follow-up system
Custom dev shop (ISK 3–5M/mo for 3-person team) Full-time AI systems architect
Total displaced: ISK 6–9M/month You get all of the above, compounding monthly

CIO's 10 Priorities — Full Map

Every item from the CIO's actionable plans is explicitly covered:

# CIO Priority Month Status
1 AI-Powered Macro Research Pipeline 1–2 Extends Mímir with Claude Opus 4.6 synthesis
2 Multi-Agent Investment Analysis 2 Three specialized agents + managed framework
3 Real-Time Sentiment & Narrative Tracking 3 Grok integration in Trading War Room
4 Investor Communications & LP Reporting 4 Automated letter drafting + attribution
5 Cross-Platform Research Integration 1 MCP + unified corpus
6 Scenario Modeling 5 Extended thinking for deep analysis
7 DD on Distribution Partners 5 Living intelligence feed on prospects
8 Thought Leadership Content 4 Content pipeline: 2–3 pieces/week
9 Proprietary Signal Generation 6 Fine-tuned models on Viska's framework
10 Meeting Preparation & Follow-Up 4 Pre/post-meeting intelligence system
Knowledge Base (not in CIO list) 2–6 Cross-cutting: all agents write to it, all queries draw from it. The institutional memory that makes everything else 10× more valuable.

Delivery Model

Aspect Structure
Duration6 months (April–October 2026)
CommitmentFull-time, dedicated
AI meetingWeekly — review what shipped, plan next week, discuss priorities
Blog postsRegular publications about the system for marketing & visibility
GitHubViska org repo with all code; Kristján (or any team member) as contributor
InfrastructureAll deployed on your infra (Supabase, Railway, Cloudflare) — you own it
At end100% handoff. System runs independently. Optional maintenance retainer.
ISK 1,200,000/month
6 months · April–October 2026
ISK 7,200,000 total · Monthly billing · 30-day exit clause

What that buys vs. alternatives

Alternative Monthly Cost Time to Value You Own It?
Hire 2 research analysts ISK 1,200,000–1,600,000 Immediate (manual work) N/A — labor, not infrastructure
Hire AI/ML engineer + onboard ISK 1,000,000–1,500,000 6–12 months Slowly
Big 4 consulting ISK 3,000,000–5,000,000 3–6 months for a report You get a PDF
Bloomberg Terminal (5 seats) ISK 1,400,000 Immediate Never — rental
AlphaSense (team license) ISK 700,000–1,000,000 Immediate Never — SaaS
This engagement ISK 1,200,000 Month 1 — extending what's live Yes — GitHub, docs, full handoff

After the 6 months

The system runs independently. Mímir transitions to a maintenance retainer (ISK 250,000/month) — same as the original Tier 2 offer. You can also bring in any developer to extend it — the codebase, documentation, and runbooks are all yours.

Why this price — Daði told Morgunblaðið that Mímir is "like having two full-time analysts." Two analysts cost ISK 1.2–1.6M/month in Iceland. This engagement costs the same as one analyst — and delivers a system that replaces three, plus a trading war room, content pipeline, CRM, and investor communications. And when I leave, the system stays.

Marketing & Visibility

Part of this engagement is public. Regular writing about what we're building — how the system works, what problems it solves, how a small fund uses AI in production. One post every 2–3 weeks. All content reviewed by Viska before publication.

For Viska

  • Positions you as a technology-forward fund — matters for LP conversations
  • "We built a custom AI intelligence department" — a differentiated story
  • Content pipeline (Month 4) produces your thought leadership — the blog posts are marketing for Viska

For boas.dev

  • Case study in production: a real fund, real results, real system
  • Demonstrates what's possible for potential clients
  • All content reviewed by Viska before publication — nothing goes out without approval

Weekly AI Meeting

Every week, one meeting. This meeting is the steering wheel. The roadmap is the plan — the weekly meeting is where reality adjusts it.

1
Demo
What shipped this week, live in the system
2
Review
Is it working? What's not landing?
3
Plan
What's next, what shifts, new ideas
4
Train
Show the team how to use and extend what's built

Risks I'm Being Honest About

Risk Mitigation
Scope creep — 10 priorities is a lot We rank them together in Week 1. Not everything ships in 6 months.
Key-person risk — system depends on me Months 4–6 explicitly include knowledge transfer. Documented, tested, handed off.
API costs — agents cost money to run Transparent tracking. Current estimate: €200–400/month for all agents. Dashboards shared.
Model changes — AI moves fast Architecture is model-agnostic. Swap Claude for GPT for Gemini without rewriting.
It doesn't work — some features underperform Monthly billing, 30-day exit. But the track record is clear: Mímir already works.

Next Steps

  1. Confirm the engagement scope and pricing
  2. Week 1: Mímir v4.0 pipeline fixes + full corpus re-ingest
  3. First weekly AI meeting scheduled
  4. Viska GitHub org created, first code pushed
  5. First blog post: "Building an AI Intelligence Department for a Hedge Fund"

The system compounds. Every month builds on the last. By October, Viska has infrastructure that most funds ten times your size don't have — and you own every line of code.