Current State
40 migrations, about 1,600 tests, 9 architecture rules.
Project
SignalClaw
SignalClaw is a parked build for reliable capture, memory, and signal infrastructure in AI-assisted work.
Capture and memory for AI-assisted work.
Active captureGoverned memoryJoan historyParked
The Problem
What problem was worth solving?
I build things to solve a specific pain, not to decorate a portfolio. This is the pressure that made this project necessary.
CRM and work systems know fragments of what happened, but AI-assisted work needs reliable capture, memory, and signal infrastructure.
SignalClaw explored that capture and memory problem through migrations, tests, architecture rules, and deal/work signal models.
The capture adapter is not yet built, so the public claim is intentionally bounded: this is a parked build, not a launched capture product.
The Build
What I built around the problem
A signal infrastructure build for AI-assisted work: capture adapters, signal detection, cross-system correlation, scoring, and delivery points. The strongest record is in architecture, migrations, tests, and rules; the capture adapter remains a known boundary.
Scoring engines
5 analysis models
Deal health (10 dimensions), deal priority (8 factors), task priority, confidence scoring, and entity matching — all deterministic and testable.
Signal types
35+ across 9 categories
Engagement drops, deal velocity, stakeholder coverage, qualification gaps, competitive signals, and more — detected from email, calendar, and CRM activity.
Data platform
16-table secure database
Deals, accounts, people, activities, signals, and scoring history — all with row-level security ensuring data isolation between customers.
Codebase
3,370+ lines · 21 modules
21 well-organized modules with strict type checking, automated quality gates, and comprehensive testing.
Reality Check
Current boundary
The point is not to pretend the work was clean. This is what the project taught, where it changed shape, or where the boundary still matters.
01
What this page should do
This page should explain why the project mattered without turning unfinished, private, or earlier work into a bigger claim than the public record supports.
02
Status
SignalClaw is currently represented as parked. The useful parts of the story are kept visible, but the page should stay clear about what is current and what is history.
03
Boundary
The public version should stay specific, plain-spoken, and careful about what is still being built.
Value
What changed because it existed
This is the practical value of the work. Where a project is unfinished, shelved, or still changing, the value is tied to what it actually taught.
Deal health scoring
10-dimension analysis
Scores deals across stakeholder engagement, qualification depth, competitive position, timeline risk, and 6 more dimensions — revealing the truth behind CRM probabilities.
Contact matching
Intelligent entity resolution
Automatically matches contacts across systems — handling name variations, title changes, and company moves so signals connect to the right people.
Smart alerting
4-stage notification management
Preference filters, quiet hours, deduplication, and daily caps prevent alert fatigue — sellers only see signals that matter.
Data honesty
Context quality assessment
The system knows when it has enough data to be confident vs. when data is thin — and tells the user, preventing false confidence.
Technical Layer
How the system is built
This is the implementation behind the work: the architecture choices, integrations, controls, and workflow decisions that made the project real enough to learn from.
Python 3.12 / FastAPI
PostgreSQL with row-level security (Azure)
Azure OpenAI (planned for LLM synthesis)
Azure Container Apps (deployment target)
Pydantic v2 strict validation
Pure-function scoring architecture (zero I/O in scoring layer)
6 enforced architecture constraints
Tiered signal processing (T1 heuristic → T2 lightweight LLM → T3 deep LLM)
Build Story
How the thinking unfolded
This is the reasoning path behind the output, not only the finished artifact.
01
Works wherever the seller works
Intelligence is extracted once and delivered to any tool: Microsoft Copilot, Teams, Dynamics, or custom dashboards. The product brings insights to where sellers already are.
02
Scoring you can trust and test
All 5 analysis engines produce deterministic, testable results. The same inputs always produce the same scores — essential for enterprise trust. Evolved from proven SalesSidekick Legacy algorithms.
03
Compliance built from day one
SOC 2 audit logging, GDPR consent management, data erasure support, and row-level customer isolation designed into the database from the first line of code — not retrofitted.
Evidence
What this project shows
This evidence is strongest when it is tied to specific work rather than broad claims.
Enterprise Data Architecture
16-table database with row-level security, customer isolation, and compliance infrastructure (audit logs, consent records, erasure support).
Intelligence Algorithm Design
5 analysis engines: 10-dimension deal health, multi-factor priority scoring, confidence assessment, and intelligent contact matching.
Quality Engineering
Comprehensive test coverage across all 21 modules. Strict type checking, automated linting, 6 quality gates, and zero-placeholder policy.
Product Architecture
Extracted and re-architected core algorithms from SalesSidekick Legacy into a standalone platform — designed to work with any CRM and any AI tool.
Continue
Want to go deeper?
Ask the Latif AI Guide about the architecture, the commercial logic, or the hard lessons behind this project.