Project Kickoff - March 2026
Argos: AI-Powered Counterfeit Detection for NAP Solutions
A new AI-powered crawling platform replacing NAP's existing Artemis scraper with intelligent pre-filtering, image-based detection, and confidence scoring - reducing manual review workload by 70–80%.
Duration
~23 Weeks
Investment
$50,820
Pilot Platforms
Amazon + Shopify
Client
NAP Solutions
Team
FT
Frances Marie Teves
Technical Project Manager
Client-facing PM, weekly updates, scope management
GQ
Gee Quidet
Chief Revenue & Solutions Officer
Accounts management, client relationship
MA
Mary Amora
Solutions Consultant
Technical solutions, architecture guidance
IN
Ian Natividad
Lead Developer
Full-stack development, AI integration, crawlers
RL
Ragan Lamoc
Developer
Backend development, build support
What is Argos?
Argos is a standalone, AI-powered counterfeit detection platform. It crawls e-commerce platforms, uses a multi-model AI engine (visual + text) to score listings for potential infringement, and presents results to your QA team for review. Fully independent infrastructure - no dependencies on legacy Artemis.
Detection Pipeline
Step 1
Configure Campaign
Keywords, images, filters, seller whitelist
→
Step 2
Crawl Platform
Amazon / Shopify via Crawlee + anti-bot
→
Step 3
AI Scoring
Multi-model: image (60%) + text (40%), per-campaign weights
→
Step 4
QA Review
Accept / reject flagged listings
→
Step 5
Export
CSV, REST API, or webhooks
Multi-Model AI Engine
Local Tier (Free)
Fuzzball
40%
Fuzzy text matching for trademark ID
CLIP Pre-Screen
Filter
ViT-B/32 embeddings for fast image triage
Heavy Tier (Vision LLM)
Deep Image Compare
60%
GPT-4o Vision / Claude - visual similarity
AI Reasoning
Score
GPT-4o / Claude - contextual analysis
Provider Abstraction Layer - config-driven routing across OpenAI, Anthropic Claude, and Google Gemini with automatic fallback
Confidence Scoring
| Score | Action |
|---|---|
| 70 – 100 | High Priority - QA review required |
| 40 – 69 | Review - Optional QA review |
| 0 – 39 | Auto-filtered - Excluded |
Tech Stack
FrontendNext.js, React, TypeScript, TailwindCSS
BackendNode.js, Express, Drizzle ORM, JWT/RBAC
DatabasePostgreSQL 16 + pgvector
QueueBullMQ + Redis
CrawlerCrawlee + Playwright
InfraGoogle Cloud Platform (Cloud Run)
Scope
In Scope (This Engagement)
- ✓ Amazon.com product listing crawling & detection
- ✓ Shopify / branded store crawling & detection
- ✓ AI-powered image + text scoring engine
- ✓ QA review dashboard & admin panel
- ✓ REST API, CSV export (ARTEMIS-compatible), webhooks
- ✓ Cross-platform matching (Amazon ↔ Shopify)
- ✓ Side-by-side validation against legacy system
- ✓ Full source code handover, docs & training
- ✓ 30-day post-handover support
Out of Scope (Evaluated, Not Included)
- ✗ Walmart, Temu, Alibaba, AliExpress crawling (future phases)
- ✗ Regional Amazon domains (.de, .co.uk, etc.) - only .com
- ✗ Smart Search / historical learning from enforcement decisions
- ✗ Dashboard & Statistics module (advanced analytics)
- ✗ Crawling Priorities & automated scheduling (manual trigger only)
- ✗ Automated enforcement actions (detection & review only)
- ✗ Legacy Artemis modifications or decommissioning
- ✗ Ongoing hosting & AI costs (passed to NAP at cost)
- ✗ Extended support beyond 30-day window (retainer available)
Users & Roles
Admin
NAP management & campaign leads
- Create & configure campaigns
- Set keywords, images, filters, whitelists
- Adjust detection thresholds per category
- Manage users and assign roles
- View analytics & performance dashboards
- Set crawling frequency & priorities
QA Reviewer
NAP's review team (~25 people)
- Review AI-flagged listings in dashboard
- Accept or reject with reasoning
- Filter by date, campaign, country, score
- Export approved results to CSV
- Decisions train the AI over time
API User Post-MVP
External systems & integrations
- Token-based API authentication
- Query detection results programmatically
- Utilize all available filters
- Integrate with existing NAP tools
- Webhook notifications for new results
User Stories
Scope & Phases
Deliverables
- Automated Amazon.com product listing crawler
- Rate-limit management & anti-bot handling
- AI-powered image comparison (multi-model: GPT-4o Vision, Claude, Gemini)
- NLP text analysis (Fuzzball fuzzy matching + CLIP pre-screening)
- Confidence scoring with AI reasoning
- Campaign configuration (keywords, images, filters, whitelist)
- Seller location filtering
- QA review dashboard with accept/reject workflow
- REST API for integration + CSV export
- Alert & notification system
Weekly Breakdown
Wk 1-2 Project setup, architecture, DB schema, GCP infra
Wk 3-4 Core backend, auth (JWT/RBAC), API endpoints, BullMQ
Wk 5-7 Frontend app, campaign UI, QA review, admin dashboard
Wk 6-8 Amazon crawler (Crawlee + Playwright)
Wk 7-9 AI scoring engine (vision + text matching)
Wk 8-9 Data export & API layer, webhooks
Wk 9-11 Testing, QA, performance, deployment
Deliverables
- Shopify store discovery and crawling engine
- Cross-platform matching (Amazon ↔ Shopify)
- Shopify-specific heuristics (store age, reviews, pricing)
- Unified dashboard across both platforms
- API User management (tokens, access control)
- Seller location: best-effort (Shopify limitation)
Weekly Breakdown
Wk 13-15 Shopify crawler, store discovery, extraction
Wk 15-16 Cross-platform matching & heuristics
Wk 16-17 Dashboard integration, testing, validation
Gate: Phase 2 begins only after Phase 1 passes acceptance testing.
Deliverables
- Side-by-side testing (min 1,000 listings/day for 10 days)
- Shadow mode → Supervised mode validation
- Complete source code handover (Git transfer)
- Architecture, API (Swagger), deployment docs
- Operations runbook & data dictionary
- Training: Ops (2x2hr), Technical (2x2hr), Platform (1x3hr)
- 30-day post-handover support
Validation Process
Week 1 Shadow Mode: Runs alongside legacy, no production actions
Week 2 Supervised: 100% human review, legacy as fallback
Sign-off Match/exceed legacy on all KPIs for 5 business days
Rollback: Legacy stays in standby 30 days. Full rollback possible within 4 hours.
Project Timeline
Task
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
Project Setup
Core Backend
Frontend App
Amazon Crawler
AI Scoring Engine
Data Export & API
Testing & Deploy
Shopify Crawler
Cross-Platform
Dashboard V2
Side-by-Side
Validation
Docs & Handover
Training
Buffer
Phase 1 (Wk 1-12)
Phase 2 (Wk 13-17)
Phase 3 (Wk 17-21) + Buffer (Wk 22-23)
Success Metrics & KPIs
≤ 5%
False Positive Rate
Adjustable per category
≤ 5%
False Negative Rate
Against known dataset
≤ 15 min
Crawl-to-Review
Up to 5,000 listings
≥ 95%
Precision
Actual counterfeits
≥ 99.5%
Uptime SLA
Production system
Phase 1 Success Criteria (Amazon)
✓ All detection KPIs met for 5+ consecutive business days
✓ Dashboard & API stable (no P1/P2 bugs for 7+ days)
✓ System processes target daily listing volume
✓ 2+ NAP team members can independently operate dashboard
✓ Miffy campaign produces actionable results
Phase 2 Success Criteria (Shopify)
✓ Phase 1 sign-off completed (prerequisite)
✓ Shopify crawler integrated and processing listings
✓ Cross-platform matching (Amazon ↔ Shopify) operational
✓ Unified dashboard across both platforms
✓ Phase 2 acceptance testing passed
Phase 3 Success Criteria (Migration)
✓ Argos matches/exceeds legacy on all KPIs (blind evaluation)
✓ Automated comparison report validates 10 business days
✓ All documentation delivered & accepted
✓ NAP team completes all 3 training tracks
✓ NAP team operates independently for 2 weeks (shadowed)
Adjustable Thresholds
All AI confidence thresholds - including image similarity, text matching sensitivity, and overall scoring - are configurable per category. No code changes required; admins can tune thresholds directly from the dashboard.
Risks & Mitigation
Platform API Changes
High Impact
Med. Prob.
Weekly monitoring of Amazon SP-API and Shopify changelogs. Weeks 22-23 buffer reserved.
Anti-Bot / CAPTCHA Evolution
High Impact
Med. Prob.
ScraperAPI + rotating proxies + CAPTCHA-solving. Proven approach from POC phase.
AI Detection Accuracy
High Impact
Low Prob.
Iterative training on NAP's labeled datasets. Adjustable thresholds. Human-in-the-loop feedback.
Third-Party Service Availability
Med. Impact
Med. Prob.
Provider-agnostic AI abstraction layer. Can swap between OpenAI, Claude, Gemini, or open-source.
Data Volume Scaling
Med. Impact
Low Prob.
Horizontal scaling on GCP (Cloud Run auto-scaling). Cost-per-listing decreases with volume.
Scope Creep
High Impact
Med. Prob.
Strict phase gates with clear entry/exit criteria. Change requests managed via sprint planning.
What We Need from NAP Solutions
Access & Credentials
- Access to existing Artemis scraper source code
- Sample datasets (known counterfeit + known legitimate listings)
- Campaign guidelines / matrix (Miffy as pilot)
- API keys or account access for testing platforms
Infrastructure
- GCP billing account for ongoing AI & hosting costs
- Decision on proxy service tier (ScraperAPI)
- Preferred communication channel (Slack, Teams, or email)
Ongoing Collaboration
- Designate primary technical contact (Ed / IT team)
- Weekly sync availability (30-60 min)
- Timely feedback on sprint demos (within 2-3 days)
- QA team available for validation phase
Sign-offs Required
- Phase 1 acceptance (before Phase 2 begins)
- Validation pass/fail criteria confirmation
- Migration sign-off (before legacy standby period)
- Final handover acceptance
Next Steps
1
Kickoff Alignment
Confirm scope, timeline, communication cadence, and team introductions. (This meeting!)
2
Access & Credentials Exchange
NAP provides Artemis source code, sample datasets, Miffy campaign matrix, and GCP billing setup.
3
Environment Setup (Wk 1)
Dev environment provisioned, CI/CD pipeline configured, Miffy test campaign data loaded.
4
First Sprint Demo (Wk 2)
Working prototype of Amazon crawler with initial detection results. Real data flowing through the system.
How We Work
Sprint Cadence
2-week sprints with a demo at the end of each. Daily async status updates keep both sides aligned between demos.
Communication
Weekly updates via email or preferred channel. Ad-hoc communication available anytime - we're reachable beyond email.
Progress Reports
Regular accomplishment reports on completed tasks, upcoming work, and blockers. Transparent sprint velocity tracking.
Escalation Protocol
Any risk impacting timeline by more than 1 week is escalated within 24 hours with a proposed mitigation plan. No surprises.
Detailed Project Timeline
23 weeks from kickoff to handover, including a 2-week buffer.
Task
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
Project Setup
Core Backend
Frontend App
Amazon Crawler
AI Scoring Engine
Data Export & API
Testing & Deploy
Shopify Crawler
Cross-Platform
Dashboard V2
Side-by-Side
Validation
Docs & Handover
Training
Buffer
Phase 1 (Wk 1-12)
Phase 2 (Wk 13-17)
Phase 3 (Wk 17-21) + Buffer (Wk 22-23)
Sprint Targets & Success Criteria
2-week sprints. Each sprint ends with a live demo to NAP stakeholders. Progress is measured by what you can see and test, not by tasks completed internally.
11 sprints · 23 weeks total
● 3 high-risk sprints (S1, S4, S6)
● 2 gate sprints requiring sign-off (S6, S9)
● 6 client dependencies flagged
PHASE 1
Amazon Crawler & AI Engine
Weeks 1–12
Sprint 1 Weeks 1–2
HIGH RISK
Goal: Prove we can stand up the full environment and ingest real data
📺 Demo Deliverable
- Live GCP environment with CI/CD pipeline running
- Database schema walkthrough (ERD on screen)
- User login flow working (admin + QA roles)
- Miffy campaign test data loaded into system
✅ Definition of Done
- Admin can log in and see empty dashboard
- CI/CD deploys to staging on git push
- Sample Miffy data visible in DB
🚨 Client Dependencies (CRITICAL)
- By Day 3: Artemis source code access - we need to understand the legacy schema to design ours
- By Day 5: Sample datasets (known counterfeit + known legitimate) - required for Sprint 4 AI training
- By Day 5: Miffy campaign matrix/guidelines - this is our pilot campaign
- By Week 1: GCP billing account linked - blocks all infrastructure provisioning
💡 PM Recommendation
Schedule a dedicated technical handoff session with Ed/IT team in Week 1. If credentials and source code aren't provided by Day 5, every downstream sprint shifts. This is the #1 schedule risk for the entire project.
Sprint 2 Weeks 3–4
Goal: Core backend is functional - users, roles, campaigns all working end-to-end
📺 Demo Deliverable
- Admin creates a Miffy campaign with keywords + reference images
- Role-based access: Admin sees everything, QA sees only review queue
- Background job queue processing visible in admin panel
✅ Definition of Done
- Admin can CRUD campaigns through the API
- JWT auth + RBAC enforced on all endpoints
- Job queue accepts crawl requests (even if crawler isn't built yet)
💡 PM Recommendation
Confirm the user roles & permissions matrix during this sprint. If NAP wants additional roles beyond Admin/QA/API, now is the time - adding roles later means reworking the auth layer.
Sprint 3 Weeks 5–6
Goal: The UI is real - NAP team can click through and give feedback
📺 Demo Deliverable
- Campaign configuration UI (keywords, images, filters, whitelist)
- QA review dashboard with mock AI-scored results
- Admin dashboard with placeholder analytics
- Seller location filtering UI
✅ Definition of Done
- NAP team can log in and navigate all screens
- Campaign creation works end-to-end (UI → API → DB)
- QA reviewer can accept/reject mock listings
🔍 Client Action Needed
UI review within 3 business days of demo. Designate 1–2 NAP reviewers for fast feedback. UI changes requested after Sprint 4 may delay the crawler integration.
Sprint 4 Weeks 7–8
HIGH RISK
Goal: First real AI-scored results on live Amazon data - the moment of truth
📺 Demo Deliverable
- Live crawl of Miffy campaign on Amazon.com
- AI-scored results with confidence scores + reasoning
- Side-by-side: Argos results vs. what Artemis would have found
- First accuracy metrics (false positive/negative rates)
✅ Definition of Done
- Crawler successfully processes 100+ Amazon listings
- AI returns confidence scores for each listing
- QA reviewer can review real results in the dashboard
- No crawler blocks/bans from Amazon for 48+ hours
⚠️ Why This Sprint Is High Risk
- Anti-bot detection - Amazon may block crawlers. ScraperAPI + proxy rotation mitigates, but first contact is unpredictable
- AI accuracy - First real-world test of the scoring engine. May need threshold tuning
- Depends on Sprint 1 data - If sample datasets were delayed, AI training quality suffers here
💡 PM Recommendation
Have NAP domain experts on standby during this sprint. When AI flags its first batch of listings, we need quick validation: “Is this actually suspicious?” Early feedback directly improves detection quality. We suggest a 30-min daily check-in during Weeks 7–8.
Sprint 5 Weeks 9–10
Goal: Full pipeline working end-to-end - crawl, score, review, export
📺 Demo Deliverable
- Complete flow: campaign → crawl → AI score → QA review → CSV export
- CSV format validated against ARTEMIS compatibility
- Alert system: notifications when crawl completes, budget thresholds
- REST API endpoints documented (Swagger)
✅ Definition of Done
- QA team can do a full mock review session (10+ listings)
- CSV export opens correctly in NAP's existing tools
- API returns data matching dashboard view
🔍 Client Action Needed
Validate the CSV export format. If the ARTEMIS-compatible format needs adjustments, this is the last sprint to make changes without impacting Phase 1 acceptance.
Sprint 6 Weeks 11–12
HIGH RISK
GATE
Goal: Phase 1 acceptance - all KPIs met, system stable, NAP signs off
📺 Demo Deliverable
- Full system demo with live Amazon data
- KPI dashboard: false positive/negative rates, crawl-to-review time, precision
- System running 5+ consecutive days with no P1/P2 bugs
- 2+ NAP team members independently operating the platform
✅ Acceptance Criteria
- False positive/negative rates ≤ 5%
- Crawl-to-review ≤ 15 min (up to 5K listings)
- Precision ≥ 95%
- Miffy campaign produces actionable results
- NAP formal sign-off
🚫 Phase Gate
Phase 2 cannot begin until Phase 1 is signed off. If KPIs are not met, we use Weeks 22–23 buffer to remediate. Milestone 2 payment (35% / $17,787) is triggered on sign-off.
💡 PM Recommendation
Start acceptance testing in Week 11, not Week 12. This gives us a full week to fix any issues before the gate. Schedule the sign-off meeting for end of Week 12 with NAP decision-makers (not just the technical team).
PHASE 2
Shopify & Cross-Platform
Weeks 13–17
Sprint 7 Weeks 13–14
Goal: Shopify crawling works - first Shopify-sourced results visible
📺 Demo Deliverable
- Shopify store discovery engine running
- First Shopify-sourced listings with AI scores
- Shopify-specific heuristics (store age, reviews, pricing anomalies)
✅ Definition of Done
- Crawler discovers and processes Shopify stores
- Results appear in QA dashboard alongside Amazon results
- Platform source clearly labeled per listing
🔍 Client Action Needed
Provide a list of known Shopify counterfeiters (if available) for validation. Also flag any Shopify-specific detection patterns your team has observed - this trains the heuristics engine.
Sprint 8 Weeks 15–16
Goal: Cross-platform intelligence - same seller detected across Amazon + Shopify
📺 Demo Deliverable
- Cross-platform matching: Amazon ↔ Shopify linked sellers
- Unified dashboard with platform filter
- API User management (token generation, access control)
✅ Definition of Done
- Multi-platform sellers flagged automatically
- Dashboard filters by platform, campaign, date, score
- Phase 2 acceptance criteria met
💡 PM Recommendation
Note on seller location for Shopify: Unlike Amazon, Shopify doesn't expose seller location reliably. We'll implement best-effort detection (WHOIS, shipping origins, payment info) but accuracy will be lower than Amazon. Set expectations with the legal team now.
Sprint 9 Weeks 17–18
GATE
Goal: Phase 2 sign-off + side-by-side validation begins
📺 Demo Deliverable
- Phase 2 acceptance demo (Shopify + cross-platform)
- Side-by-side comparison dashboard: Argos vs. legacy Artemis
- Shadow mode activated - Argos running alongside legacy
✅ Definition of Done
- Phase 2 signed off by NAP
- Shadow mode processing 1,000+ listings/day
- Automated comparison report generating daily
🚨 Client Dependencies
QA team (~25 people) must be available for parallel reviews starting this sprint. Legacy Artemis must remain running. Milestone 3 payment (25% / $12,705) triggered on Phase 2 sign-off.
PHASE 3
Validation & Handover
Weeks 17–21 + Buffer (22–23)
Sprint 10 Weeks 19–20
Goal: Argos proven to match or exceed legacy - docs and training complete
📺 Demo Deliverable
- 10-day blind evaluation results: Argos vs. legacy on all KPIs
- Supervised mode: 100% human review with legacy as fallback
- Architecture docs, API (Swagger), deployment runbook
- Training sessions: Ops (2×2hr), Technical (2×2hr), Platform (1×3hr)
✅ Definition of Done
- Argos matches/exceeds legacy on all KPIs for 5+ business days
- All documentation delivered and accepted
- NAP team completes all 3 training tracks
- NAP team operates independently (shadowed by Symph)
🔍 Client Action Needed
Schedule training sessions now. 11 hours total across 3 tracks - needs calendar coordination with ops team, IT team, and platform users. NAP must designate who attends each track.
Sprint 11 Weeks 21–23
FINAL
Goal: Clean handover - NAP owns the system, Symph transitions to support
📺 Demo Deliverable
- Complete source code transferred (Git repo handover)
- Operations runbook & data dictionary
- NAP team running system independently for 2+ weeks
- 30-day post-handover support period begins
✅ Final Acceptance Criteria
- All code, docs, credentials transferred
- NAP confirms independent operation capability
- Legacy standby period begins (30 days, rollback within 4hr)
- Final sign-off from NAP
🎉 Milestone 4 Payment
Final 20% ($10,164) triggered on sign-off. Weeks 22–23 serve as buffer for any remaining items. Post-handover support (30 days) is included - extended support available via retainer.
💡 PM Recommendation
Discuss post-handover retainer options during this sprint. 30 days of support goes fast - if NAP wants ongoing AI model tuning, crawler maintenance, or feature work, scope it before the support window ends.
Weekly Updates
Weekly status reports will appear here as the project progresses.
Investment & Payment Schedule
Milestone-based payments. Each requires written acceptance.
20%
Milestone 1
Project Kickoff
$10,164
35%
Milestone 2
Phase 1 Delivered
$17,787
25%
Milestone 3
Phase 2 Delivered
$12,705
20%
Milestone 4
Phase 3 Complete + Sign-off
$10,164
Total Project Investment:
$50,820
Ongoing Costs (Post-Launch)
AI Inference Costs (monthly)
| Volume | Estimated Cost |
|---|---|
| 10,000 listings | $120 - $180 |
| 50,000 listings | $450 - $650 |
| 100,000 listings | $800 - $1,100 |
| 500,000+ listings | $3,200 - $4,500 |
Per-listing cost decreases with volume. Budget caps & alerts included.
Infrastructure (monthly, at cost)
| Service | Estimated Cost |
|---|---|
| Compute (Cloud Run/GKE) | $30 - $80 |
| Database (PostgreSQL) | $10 - $40 |
| Storage (GCS) | $5 - $20 |
| Hosting subtotal | $45 - $140/mo |
| ScraperAPI (proxy, separate) | ~$49/mo |
Documents
Project documents, links, and shared resources will be organized here.