Home Projects Career Skills Blog Resume Contact

Work

Built in production. Measured in outcomes.

Enterprise AI systems, ERP architecture, data platforms, and automation — designed for real organizations with real constraints.

⭐ Flagship Project

Enterprise AI Operating System

The most ambitious thing I've built: a complete AI operating system for a multi-company industrial enterprise — from ground zero. 101 specialized agents, 10 MCP servers, 4 production web apps, and a 3-layer RAG knowledge system — all live, all interconnected, all running today.

101+AI Agents
10MCP Servers
4Production Apps
329%Documented ROI
160+Automated Workflows
2.55M+RAG Vectors
Agentic AI Multi-agent Orchestration LLMOps MCP Servers RAG Systems Context Engineering Next.js 16 Odoo v17 Docker Grafana / Loki
What Was Inherited
  • Critically unstable production systems with no monitoring
  • Dozens of manual hours consumed by repetitive reporting
  • Virtually no documentation — just 3 SOPs total
  • Zero AI, zero automation, zero integration between systems
What Was Built (5 months)
  • 101 AI agents across 9 departments using BMAD methodology
  • 10 MCP servers — agents connected to SAP B1, Odoo, Jira, GitHub, Loki, Beta system
  • 3-layer RAG: Qdrant (2.55M+ vectors) + pgvector (HNSW, 22ms) + Gemini Vision
  • 4 production web apps shipped in 3 months
  • 19 Docker services, 99.7% uptime, full Grafana/Prometheus/Loki observability
  • 6 custom Odoo v17 modules, autonomous SAP/Beta/Exotel data pipelines
Results
  • Production environment: critically unstable → near-zero errors
  • System health transformed from degraded to high-performing
  • Development throughput more than tripled
  • SOPs: near-zero → comprehensive coverage across all departments
  • 329% documented ROI — 12-month roadmap now executing Q2
  • 120+ hours/month freed through hyperautomation
Stack
  • Claude Opus / Sonnet / Haiku, Gemini Vision API
  • Qdrant, pgvector, Neo4j, Redis — multi-layer RAG
  • Next.js 16, Drizzle ORM, Supabase, PostgreSQL
  • Odoo v17, SAP B1 (OData), N8N, Exotel, Brevo
  • Docker, Traefik, Grafana, Prometheus, Loki, cAdvisor
Live App ● Live
🎯 72,712 AI-scored matches

Radar Y — Project Intelligence Platform

Scans construction project data across India, scores customer-project fit using AI, and surfaces the highest-priority opportunities for the sales team — automatically.

Next.js 16Drizzle ORMpgvector20 AI agents
What It Does

Tracks 2,529 construction projects across India. 20 AI agents score customer-project fit. Surfaces pitch-ready intelligence: stakeholder maps, project timelines, zone-level opportunity scoring. Went from V1 → V7 in 3 weeks.

Scale
  • 2,529 projects tracked, 6,612 stakeholders mapped
  • 72,712 AI-scored customer-project match combinations
  • 22 pages, 36 API routes, 93 automated tests
Live App ● Live
🧠 511 customers × 127 fields

Customer Intelligence & Enrichment Engine

Transforms sparse CRM records into deeply profiled customer intelligence — lead scoring, churn prediction, ghost-phase detection — using 27 AI agents running in parallel pipelines.

Next.jsSupabase27 AI agentsJob queues
What It Does

Pulls raw customer data from SAP, Odoo, and Beta systems. 27 agents enrich each record across 127 fields — RFM segmentation, lead scoring, ghost-phase detection, churn prediction, news classification. 20 database-driven job queues with retry and dead-letter handling.

Scale
  • 511 customers fully profiled (62,619 data points total)
  • 6,002 leads in pipeline, 1,897 news intelligence signals
  • 21 pages, 61 API routes, 31 database tables
Live App ● Live
👁️ 92-95% vision accuracy

AI Safety Inspector

Uses Gemini Vision to analyze equipment photos and safety inspection documents — extracting compliance data, flagging risks, and generating structured reports automatically.

Gemini VisionNext.js 16Drizzle ORM
What It Does

Equipment inspectors upload photos and documents. Gemini Vision extracts structured data — equipment condition, safety compliance status, risk flags — at 92-95% accuracy. Generates inspection reports automatically, stores results in a queryable database.

Why It Matters

Manual safety inspection documentation was slow, inconsistent, and hard to query. Now it's fast, structured, and searchable — inspectors focus on the inspection, AI handles the paperwork.

AI & LLMOps Completed
🗂️ 2.55M+ vectors · 22ms retrieval

Enterprise RAG Knowledge System

3-layer retrieval architecture that gives the entire enterprise a searchable, AI-queryable memory — equipment knowledge, safety data, operational docs, all surfaced in under a second.

QdrantpgvectorRedisGemini VisionHNSW
Architecture
  • Layer 1: Qdrant — 2.55M+ vectors, 5 collections, semantic search
  • Layer 2: pgvector with HNSW indexing — 150ms raw → 22ms with Redis cache (98% hit rate)
  • Layer 3: Gemini Vision — equipment images and safety documents at 92-95% extraction accuracy
Impact

Organizational knowledge that previously required asking the right person now surfaces in under a second. Non-technical staff can query enterprise data in plain language.

DevOps & Data Completed
🔗 10+ systems unified

Multi-System Integration Architecture

Designed and built the integration layer connecting SAP B1, Odoo, Beta (152 endpoints), Exotel, Brevo, Jira, GitHub, and more — eliminating hundreds of daily manual data transfers.

OData APIREST/SOAPWebhooksRedis queuesPython
Pipelines Built
  • SAP B1 ↔ Enrichment (OData, 500 records/page, daily autonomous sync)
  • Beta ↔ Enrichment (MySQL/AWS RDS, 2-hour polling)
  • Odoo ↔ Enrichment (XML-RPC, 4-hour lead sync)
  • Brevo ↔ Odoo (5-script dedup pipeline: identify → pre-merge → auto-merge)
  • Exotel ↔ Odoo (webhooks, real-time, DLT compliance)
Impact

Data that used to flow via WhatsApp messages and spreadsheets now moves autonomously, reliably, and with full audit trails.

ERP Completed
🏢 3 companies · 6 custom modules

Multi-Company Odoo v17 ERP Architecture

Full Odoo v17 implementation unifying manufacturing, equipment rental, and distribution — 6 custom modules, 75 security groups, 240+ ACLs, autonomous data pipelines in and out.

Odoo v17PythonPostgreSQLXMLMulti-company
Custom Modules
  • ygroup_security — 75 groups, 90+ record rules, 240+ ACLs, multi-company RBAC
  • exotel_integration — click-to-call, DND/NDNC, campaign automation
  • ar_auto_assignment — automated AR executive routing
  • unbill_chase_team — unbilled order chasing automation
  • plm — BOM approval workflow
  • website_ygroup — corporate site with CRM-integrated forms
DevOps & Data ● Live
📊 19 services · 99.7% uptime

Production Infrastructure & Observability Stack

Built and runs the entire production infrastructure — 19 Docker services, full Grafana/Prometheus/Loki observability, Traefik reverse proxy, automated SSL, alerting, and backups.

DockerTraefikGrafanaPrometheusLokiRedis
Stack
  • 19 Docker services on Hostinger KVM 8 (Ubuntu 24.04, 400GB)
  • 4 PostgreSQL databases: Supabase, pgvector, LOKI, Odoo
  • 7 Grafana dashboards, Prometheus 15-day retention, 50M+ Loki log entries
  • Traefik with automatic Let's Encrypt HTTPS across all apps
  • Redis 7 — 98% cache hit rate, LRU eviction, AOF persistence
  • Automated backups: 21 cron jobs including nightly DB dumps
Automation ● Live
⚡ 160+ workflows · 9 departments

Enterprise Hyperautomation Platform

160+ automated workflows spanning every department — from AR chasing and lead sync to compliance monitoring and daily intelligence briefings. 120+ hours of manual work eliminated monthly.

N8NPythoncronJob queuesWebhooks
Automation Layers
  • 21 cron-scheduled AI agent jobs (health checks, AR aging, dispatch, e-way bill, standups, board packs)
  • 20 database-driven job queues with retry + dead-letter handling
  • 78 enrichment pipeline scripts
  • 3 launchd daemons: daily health, weekly reports, monthly board prep
Coverage

CEO briefings, Sales pipeline, Operations dispatch, Finance AR, Manufacturing WIP, Safety training, IT monitoring, HR workflows, Marketing lead generation.

ERP Completed
🏢 30+ OCA modules configured

HRMS Migration & Foundation

Designed the full HR tech stack on Odoo using 30+ OCA modules — covering the complete employee lifecycle from recruitment through payroll, with clean upgrade paths and no customization debt.

Odoo v17OCA modulesPythonPostgreSQL
Scope

Recruitment, onboarding, attendance, leave management, contracts, payroll computation, performance evaluation — all unified. Dependency graph designed for clean upgrades and minimal future tech debt.

Security Completed
🛡️ Critical vulnerabilities remediated

Security Audit & DPDP Act Compliance

Comprehensive security audit uncovering critical vulnerabilities in a legacy platform. Designed a 4-phase remediation strategy and full DPDP Act 2025 compliance roadmap.

Security AuditSQL InjectionDPDP Act 2025Risk Modeling
What Was Found
  • SQL injection exposure across hundreds of query patterns
  • SSL verification disabled in multiple system locations
  • No compliance framework for India's DPDP Act 2025
What Was Built

4-phase remediation plan with prioritized execution. DPDP compliance roadmap: consent management, data minimization, breach notification, cross-border transfer controls. Security elevated from reactive to proactive governance.

AI & LLMOps Completed
📞 Significant telephony cost reduction

AI Voice Agent Platform

Built an AI-powered voice agent platform on Gemini + Exotel — replacing legacy telephony with intelligent call handling, real-time transcription, sentiment analysis, and automated CRM sync.

Gemini APIExotelN8NPostgreSQLPython
What It Does

Inbound call handling via AI agents, real-time transcription, post-call sentiment analysis, automated Odoo CRM sync. Collections, customer service, and sales calls all AI-assisted. Scalable architecture supporting multiple concurrent voice agents.

Automation Completed
📘 3 → comprehensive SOPs across all depts

Interactive SOP Playbook System

Replaced static, ignored documents with interactive HTML5 SOPs — visual process flows, architectural diagrams, step-by-step guides. Living documentation that actually gets used.

HTML5CSS3JavaScriptSVG Diagrams
What Changed

SOPs went from near-zero to comprehensive coverage across all departments. 96+ architectural diagrams. Each SOP is a self-contained web app — searchable, visual, offline-capable. Process changes update the doc, not a PDF nobody reads.