Enterprise Analytics · Production·Sole Architect & Developer·2024 — PresentProduction

ARGUS

ARGUS cover

Named after the all-seeing giant from Greek mythology. Multi-level drill-down analytics, sales pipeline tracking, client portals, and a presentation mode with a canvas drawing overlay for live client meetings — all hand-built. 36K lines of TypeScript, 57 REST routes, 99%+ uptime.

Scale
108+
Clients managed
~480
Active campaigns
10,000+
Prospects tracked
166
Active users
57
REST endpoints
~36,000
TypeScript lines
~50ms
Response p50
~200ms
Response p99
99%+
Uptime
Stack
Next.js 16App RouterEdge MiddlewareJWTLRU CacheRaw SQLCTEsPrometheusTypeScript
Highlights
01

Multi-level Drill-down Analytics

Client → Account → Campaign with 15+ computed metrics per row. One click to go from org-level overview to individual campaign performance. Raw SQL with CTEs and window functions for sub-200ms response times.

02

Sales Pipeline Tracking

CRM stats, Calendly bookings, call volume trends, and insight reports with JSONB snapshots. Real-time visibility into every stage of the sales funnel — built for a team managing 480+ concurrent campaigns.

03

Client Portal

External authentication so clients access their own data without any internal involvement. Scoped JWT tokens, RBAC via Edge Middleware, read-only views that refresh on a tiered cache schedule.

04

Presentation Mode

HTML5 Canvas drawing overlay, laser pointer, and row emphasis for live client meetings. Built because screen-sharing spreadsheets was embarrassing. Now the primary tool for weekly client reviews.

05

Tiered LRU Cache

In-memory LRU cache with tiered TTLs ranging from 1 minute to 5 hours depending on data volatility. Pattern-based invalidation on writes. 80%+ cache hit rates — the difference between 50ms and 3s at scale.

06

Fuzzy Name Matching

Calendly attribution required matching 166 users across 100+ nickname variations. Exact matching hit 60% accuracy. Fuzzy matching with a curated nickname map pushed it to 95% — critical for commission reporting.

What Broke & How I Fixed It
Problem

Dashboard with 30 metrics → nobody used it

Fix

Narrowed to 5 actionable KPIs → usage went from 10% to 80%

Problem

Exact name matching → 60% accuracy on Calendly attribution

Fix

Fuzzy matching with 100+ nickname variations → 95% accuracy

Problem

Aggressive caching without invalidation → stale data everywhere

Fix

Pattern-based cache invalidation on every write → always fresh

Problem

N+1 queries in drill-down → 3s load times

Fix

Collapsed to single CTE query → 200ms

© 2026 Subin Joshua Sunil · Built in the dark.

← Back to portfolio