Case studies

Ten anonymized engagements. Problem, role, technology, timeline, and result for each.

Representative anonymized engagements. Client references and audit evidence available under NDA during procurement.

Banking · Canada-92%time to audit a decision

Problem: After regulator pressure on agent transparency, the bank could not reconstruct why a retention offer had been made.
Wouessi role: Architected and shipped a replayable retention agent over the existing CRM; built the Replay Envelope (Context Object + Model State + Policy State + Environment State).
Tech: Python · LangChain · Postgres · pgvector · MCP · Azure OpenAI
Timeline: 20 weeks · 2 senior engineers · 1 delivery lead
Result: Compliance now reconstructs any agent action in under 2 minutes (was 4 weeks). 8,000 weekly conversations covered.
Why it mattered: Compliance can now answer regulator questions in minutes instead of weeks, and the bank can defend every retention offer with the model card, prompt version, and data context that produced it.

Public sector · Canada100%Protected B residency

Problem: Program officers needed AI assistance for briefing notes, but no PHI or policy data could leave the federal perimeter.
Wouessi role: Sovereign on-prem deployment with self-hosted Llama 3.1 70B and a sovereign Context Object store.
Tech: Llama 3.1 · Kubernetes on-prem · pgvector · MCP · Azure ATIP
Timeline: 26 weeks · 3 engineers · 1 governance lead
Result: Zero data left the perimeter. ATIP requests now answered with one-click replay. Director of Digital signed off on launch.
Why it mattered: Program officers got the speed of AI without federal data ever leaving the sovereignty boundary, and the Director of Digital has hand-back evidence that survives any audit.

Healthcare · USA18 wkship to live audit

Problem: CMIO needed an AI care-navigation agent that survived clinical ethics review and HIPAA scrutiny.
Wouessi role: Built a refusal-tuned care-navigation agent with per-patient audit trail and full Replay Envelope.
Tech: AWS · Bedrock · DynamoDB · MCP · HIPAA-aligned controls
Timeline: 18 weeks · 2 senior engineers · 1 clinical informatics partner
Result: CMIO sign-off on day 90. Clinical ethics review passed at first attempt. Live with 12 specialty clinics.
Why it mattered: Clinical staff trust the agent because every recommendation is traceable to its source, and patients get faster routing without losing the human review on anything that matters.

Retail · Canada-70%time to publish offers

Problem: Six-vendor stack with quarterly publishing cycles; loyalty team wanted to publish weekly.
Wouessi role: Replaced six vendors with one bilingual loyalty portal and a workflow engine.
Tech: Next.js · Postgres · Redis · Hasura · Stripe
Timeline: 32 weeks · 4 engineers · 1 product designer
Result: Cross-channel offers now publish in 4 hours (was 14 days). 2.6M members served in EN+FR.
Why it mattered: The marketing team went from quarterly to weekly campaign cycles without losing the bilingual parity their members expect, and procurement can audit every promotion against the data it was built from.

Insurance · Canada4 wkunderwriter capacity returned

Problem: Claims triage was bottlenecked behind manual document review.
Wouessi role: Built a triage AI with confidence-graded output and refusal-on-uncertainty defaults.
Tech: Python · OpenAI · Postgres · ServiceNow integration
Timeline: 12 weeks · 2 engineers · 1 underwriting SME
Result: 4 weeks of underwriter capacity returned in the first quarter. Refusal rate kept above 35% on edge cases.
Why it mattered: The underwriter team got four full weeks of capacity back in Q1, which paid for the engagement before the second quarter began.

AI consulting · USA6 wkboard deck to working agent

Problem: Board approved an AI copilot pilot; team had no prior production AI experience.
Wouessi role: Crash-build of a working agent with full evaluation harness, in 6 weeks.
Tech: OpenAI · LangChain · Pinecone · TypeScript
Timeline: 6 weeks · 1 senior engineer · 1 product partner
Result: First working agent at week 6. Delivered eval harness, model card, and runbook the customer's team could continue.
Why it mattered: The board saw a working agent on schedule, and the customer’s team owns the eval harness, model card, and runbook to keep iterating without us.

Banking · Canada100%WCAG 2.2 AA pass

Problem: Public-facing banking site flagged for accessibility violations.
Wouessi role: Manual + automated audit, remediation playbook, on-team training.
Tech: axe-core · Lighthouse · NVDA · VoiceOver · JAWS
Timeline: 9 weeks · 2 accessibility specialists
Result: 1,200-page site reached WCAG 2.2 AA pass. Internal team certified to maintain it.
Why it mattered: The bank’s public banking site is now defensible under AODA and Section 508 scrutiny, and the internal team is certified to maintain the standard without external help.

AI consulting · Canada3 banksreference accounts

Problem: Three Canadian banks wanted a sovereign AI governance pilot before Stanza-46 GA.
Wouessi role: Bespoke replay layer over their existing AI deployments. Pilot is the precursor to the Stanza-46 product.
Tech: Python · custom Replay Envelope · Postgres
Timeline: Ongoing · early-access program
Result: Three signed pilots ahead of Stanza-46 v0.1 release.
Why it mattered: Three banks have a reference implementation of governed AI months before Stanza-46 GA, and Wouessi has live production data to harden the product against.

Nonprofit · Canada-58%intake processing time

Problem: Front-line staff handled intake, vet records, and adoption matching across three disconnected spreadsheets and a paper binder.
Wouessi role: Designed and shipped a unified intake and adoption portal with role-based access for staff, vets, and foster families. Bilingual where the donor and grant audience required it.
Tech: Next.js · Postgres · Auth.js · S3 storage · printable PDF receipts
Timeline: 10 weeks · 1 senior engineer · 1 product partner
Result: Intake processing time cut by 58 percent. Adoption matching window from days to hours. Donor and grant reporting now exports in two clicks.
Why it mattered: Front-line staff stopped losing intake records to spreadsheet sprawl, and donor reporting that used to take two days now exports in two clicks, freeing the team to spend their time on the animals, not the paperwork.