Dheerendra Rahul

Sales team works 9:30–6, half their time goes to qualifying leads instead of closing them. People leave; training cycles eat into output. Lead-to-cash takes too long and too much human glue.

• Marketing Agent — automated campaigns on Meta (Instagram + Facebook), WhatsApp, Google Ads, YouTube; auto-boosts reels and ads
• Lead Capture & CRM Sync — pulls leads from IndiaMART and Zoho on a schedule; deduplicates and normalises
• Lead Qualifying Agent — WhatsApp AI conversation with real-time product/pricing data; only warm leads escalate to humans
• Callback Voice Agent — calls back qualified leads, walks pricing, captures objections; every call transcribed and summarised by a second LLM
• Counter-Offer Engine — extends a pre-approved counter-offer if buyer hesitates; final logic is deterministic, not LLM-decided, to protect margin
• Bulk-Deal Closure Agent — for B2B bulk orders, computes the best-allowed counter-offer; escalates with full context
• Payment Link Generator — generates a payment link only after manual amount-verification
• Logistics Agent — gate-camera image, barcode scan, vehicle/porter photo, way-bill via Tally
• Live Tracking Agent — Porter live-tracking link sent to buyer; closes the loop on delivery

• Sales team focuses only on high-intent leads — qualifying time dropped to near zero
• 24/7 lead engagement (was 9:30–6 only); follow-up latency went from hours to seconds on WhatsApp
• Counter-offer logic kept deterministic so margin is preserved even when LLMs are creative
• Logistics handoff fully traced — image, barcode, way-bill, tracking link — with no manual data entry

40-microservice platform (Java/Go/Node, 500+ endpoints) burning $3.5–5.5K/mo to serve a real-world userbase of <5K. Long deployment cycles, test pyramid impossible to keep green, multiple databases (Neo4j, Postgres, Elasticsearch, MySQL), HIPAA gaps, no clear B2B story.

• Right-sized to actual usage (10–15K users target) — collapsed 40 services into a single FastAPI modular monolith with preserved service boundaries for future re-extraction
• Replaced heavy components with in-process Python equivalents — Auth0 → JWT+bcrypt+pyotp, Elasticsearch → PostgreSQL tsvector + pg_trgm, Redis → TTLCache, Kafka → asyncio.Queue + APScheduler, Neo4j → PostgreSQL recursive CTEs
• Removed Ghost CMS — moved to S3 + CloudFront for CMS delivery
• Authored Phase 7 HIPAA / SOC 2 plan: PostgreSQL RLS, field-level PHI encryption via AWS KMS, SAML/SCIM federation, external pen-test, vendor security review
• Converted to a B2B multi-tenant platform: 45+ per-organisation feature flags, 7-role / 58-permission RBAC, per-tenant theming and white-labeling, on-the-go onboarding
• Internationalisation including Arabic right-to-left layout
• Delivered using AI-accelerated / spec-based development — compressed a projected 1–3 month rewrite into focused multi-week sprints

• Infrastructure cost: $3.5–5.5K/mo → $340–545/mo (85–90% reduction)
• Engineering team needed to run the platform: 4–8 → 1–2 (75% reduction)
• Deployment time: hours → minutes (single container vs 40)
• Platform now sellable to organisations as a white-labeled B2B product

Legacy ERP shipped as a Windows installer; on-site DBA needed for every release; integration partners shared a single cookie-based session; oversized cookie containing every permission breaking the app; old VB.NET code with no documentation; client install took hours.

• Auth split for integration partners — ROPC flow for service-to-service; web-app sessions retained for human users; partners no longer need browser cookies
• AI-driven legacy → modern migration — VS Code plugin auto-documents legacy VB.NET; "Code Gini" agent emits .NET 8 + React equivalents; module-by-module parallel run until parity
• React performance fix — Axios + Redux replacing fetch sprawl; Vite for fast dev cycles; coarse-grained API replaced with focused endpoints
• Single-script installer + auto-config — Python+WiX builds and ships a fresh installer per run; auto-uninstalls old, restores config, opens admin UI on first start
• Database-change tracking & release notes — AI+Python tool diffs schema between releases; output flows into release notes automatically — DBA dependency removed
• JWT migration to fix cookie-bloat — cookie permissions moved to server-side RBAC; cookie shrinks to a single token; permission size no longer a request-size constraint

India-only travel booking platform (Razorpay, INR, English-IN, single domain). Wanted to expand to UK, US, Singapore, UAE on the same domain with country-specific content, pricing, payments, and SEO — without forking the codebase.

• URL pattern offpeakbreak.com/{in,uk,us,sg,ae}/ — single SSL cert, single deploy, country path prefix
• GeoIP detection at CloudFront edge (free, accurate); cookie + manual switcher always available
• Country-wise multi-tenancy at the data layer — added country_code to all customer-facing tables; JPA aspect injects the WHERE clause automatically
• Country and CountryConfig tables — adding a new country becomes a config change, not code
• Payment provider abstraction — Razorpay for India, Stripe for everyone else, future providers plug in cleanly
• User-identity and PII segregated in the schema — paves the way for in-country data residency without code changes
• i18n with react-i18next and per-country English variants (en-IN, en-GB, en-US, en-SG, en-AE); hreflang tags for proper Google country-targeting

• 5 countries live in 6–8 calendar weeks at ~10 person-weeks of engineering
• Forward-compatible migration — India site behaves identically until the multi-country flag is enabled
• Per-country data residency now achievable without downtime if compliance demands it later
• Per-country SEO ranking properties on Google Search Console from day one