Production AI · 2025–2026

Case Studies

End-to-end AI solutions, driving results in production.

Financial Services / Private Credit · Case 01
GI Bletchley

Mapping Credit Risk Across 14,171 Borrowers in Real Time

Problem

Private credit lenders and PE credit investors had no independent, real-time reference for consensus loan pricing across BDC portfolios. Valuation divergences — like one lender marking a borrower at 28.7¢ while another marked it at 99.9¢ — were invisible until it was too late.

Solution

GIA built GI Bletchley, a production AI surveillance platform that ingests structured EDGAR filings across 150 BDCs, normalises borrower-level data, and computes the Bletchley Consensus Mark — a weighted consensus price across all reporting lenders for each borrower. The SIMONS pattern engine flags divergences and deterioration signals in real time.

Models
Claude Sonnet GPT-4o
Stack
Next.js Supabase Tailwind shadcn/ui
0114,171 — borrowers covered
02150 — BDCs tracked
03CRITICAL — status alerts delivered before market consensus catches up
04$500K+/yr — institutional licensing
GI Bletchley · Surveillance Console Live
CRITICAL Tessera Holdings
28.7¢ ↔ 99.9¢
Bletchley Consensus Mark87.4¢
Reporting BDCs150
24h flagged borrowers42
Private Credit / Workflow Automation · Case 02
VALIS

Automating Private Credit Origination from Sourcing to IC Memo

Problem

Private credit origination is labour-intensive. Sourcing targets, gathering financials, running credit analysis, and producing investment committee memos each require specialist time that does not scale.

Solution

GIA built VALIS — an AI-native origination platform that identifies target borrowers, enriches data from Companies House and SEC EDGAR, scores credit quality, and generates full IC memos automatically. The system covers UK and US markets and integrates outreach automation for high-volume prospecting.

Stack
Next.js Supabase FastAPI SmartLead
01Weeks → hours — origination cycle
02Standardised — IC memo quality and consistency
03Fund I — deployed on GIA’s live pipeline
VALIS · Origination Workflow Fund I
Stage 1 · Source
Identify target borrowers from CH & EDGAR
Stage 2 · Enrich
Pull financials & company data
Stage 3 · Score
AI credit quality assessment
Stage 4 · IC Memo
Generating investment committee memo
Stage 5 · Outreach
Queued · SmartLead integration
Financial Services / Agentic AI · Case 03
ATLAS Agents

A 24-Agent AI Trading Swarm with Evolutionary Optimisation

Problem

Systematic trading strategies degrade as market regimes shift. Manual strategy review is slow and biased. Backtesting infrastructure is fragmented across data sources and execution environments.

Solution

GIA built ATLAS — a 24-agent AI trading swarm deployed on Azure. The SIMONS pattern engine identifies 31 confirmed mean-reversion patterns. The JANUS meta-layer detects regime shifts and reweights the agent cohort dynamically. Darwin v3 uses evolutionary optimisation to continuously improve agent prompts without human intervention.

Stack
Azure Python Databento Tradovate IBKR
01+29.6% — backtested return
020.74 — Sharpe ratio
03750,000+ — organic views on launch
041,800 — GitHub stars
05$499/mo — first paying marketplace customer
ATLAS · 24-Agent Swarm Azure
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
Mean-reversion patterns31 confirmed
JANUS regimeLow-vol · risk-on
Backtested return+29.6%
Sharpe ratio0.74
Private Credit / Pattern Detection · Case 04
SIMONS-Credit

35 Validated Credit Patterns — 78% Precision, 4.15× Lift

Problem

Early warning signals in private credit portfolios are buried in unstructured BDC filings. No systematic approach existed for pattern detection at scale across the market.

Solution

GIA built the SIMONS-CREDIT pattern engine — trained on 1,159 clean loans across 22 sectors from SEC EDGAR BDC filings. 35 patterns validated. The engine identifies divergence signals between lenders marking the same borrower at materially different prices, flagging CRITICAL status before the market reprices.

Stack
Python Supabase Next.js
0135 — validated patterns
0278% — precision on lead indicator
034.15× — information lift over baseline
04Backbone — of GI Bletchley’s institutional monitoring product
SIMONS-Credit · Pattern Library 35 validated
P-017 Mark divergence ≥ 40¢
84%
P-009 Lender concentration shift
79%
P-022 Sequential mark-downs
76%
P-031 Sector dispersion spike
72%
P-014 NAV deviation > 3σ
68%
Information lift vs. baseline 4.15×
Publishing / Editorial Workflow · Case 05
Village to Valley

An 11-Agent AI Editorial Pipeline for Long-Form Publishing

Problem

A 70,000-word political memoir required substantive editorial development across 11 specialist functions — structural analysis, fact-checking, tone calibration, narrative coherence, political sensitivity review. Traditional editorial processes take months and cost six figures.

Solution

GIA built an 11-agent AI editorial pipeline — each agent a specialist running sequentially and in parallel across the manuscript. Agents cover structural editing, fact-checking, tone analysis, political sensitivity, pacing, and consistency. The human author retains full creative control; agents handle the analytical load.

Stack
Claude Code Claude Sonnet Python
01Hours — full editorial pipeline runtime (vs. months)
0275,000 — words · target manuscript
03Ascalon — publishing under Ascalon Publishing imprint
Village to Valley · Editorial Pipeline 11 agents
Structural Editor
Complete
Fact Checker
Complete
Tone Analyst
Complete
Political Sensitivity
Complete
Pacing & Cadence
Complete
Narrative Coherence
Complete
Consistency Pass
Complete
Citation Verifier
Running
Style Continuity
Running
Voice Calibration
Running
Final Review
Queued

Get AI into production.

Get Started