Agentic AI Suite Portfolio — India

Challenge

Outbound sales teams were burning hours on manual prospecting — research, lead scoring, and outreach list-building were slow, inconsistent, and impossible to scale without adding headcount.

Approach

We built a multi-agent pipeline (Searcher, Analyst, Enrichment) on LangGraph that browses the web for buying signals, scores decision-makers, and validates contact data, with a human-in-the-loop approval gate before any list ships — running as a Dockerized microservice portable across AWS and GCP.

Results

• Replaces the manual research workload of roughly 10 SDRs
• Cuts SDR research time by 60%
• Generates a validated, outreach-ready lead list in under 60 seconds

Challenge

Running consistent campaigns across five channels — Email, WhatsApp, Instagram, LinkedIn, Google Ads — from a single brief meant either slow manual adaptation per platform or brand voice drifting between channels.

Approach

A Writer → Reviewer → Publisher agent swarm takes one product brief and produces platform-native creative for all five channels simultaneously, with predictive tone tuning per platform and a brand-consistency score tracked on every campaign before a human signs off on publishing.

Results

• One brief becomes 5 platform-native campaigns simultaneously
• 100% brand-voice consistency tracked across every channel
• Human-in-the-loop review gate before any campaign goes live

Challenge

D2C and marketplace sellers had ad spend, inventory, and margin data sitting in disconnected systems, so by the time a margin problem showed up in a weekly report, the budget behind it was already spent.

Approach

We connected ad platforms (Google, Meta, GA4), marketplaces (Shopify, Amazon, Flipkart) and inventory/ERP data into one semantic layer, with an XGBoost-driven control loop that autonomously pauses underperforming ads and triggers re-orders the moment ROAS crosses a set threshold — no human required in the loop.

Results

• Autonomous action when ROAS drops more than 15% — ads paused and budget reallocated instantly
• Drift detection triggers model auto-retraining the moment performance degrades
• Live margin and spend observability protecting roughly 30% more margin

Challenge

Delivering genuinely one-on-one tutoring quality to a large student base isn't possible with human tutors alone, and most ed-tech chat tools either hallucinate answers or feel generic.

Approach

We built a knowledge-graph-backed RAG pipeline with hybrid keyword and semantic search plus a reranking layer, grounding every answer in the actual course material, then layered dynamic difficulty adjustment based on each student's real performance.

Results

• Serving 1,000+ tutoring sessions at sub-200ms response latency
• Knowledge-graph maps true topic dependencies to pinpoint each student's actual gaps
• Fully containerised MLOps stack — CI/CD, drift detection and observability run with zero manual touch

Challenge

A single broken Google Tag Manager trigger can silently corrupt attribution for weeks before anyone notices — most teams only find out once monthly numbers stop reconciling.

Approach

We built an agent that runs a full one-click audit of a GTM container, flags broken tags and missing triggers, and suggests (or auto-applies) the fix, alongside a standing GDPR/CCPA consent-mode compliance check.

Results

• Full container audit runs in minutes instead of the multi-day manual process it replaces
• Consistently high data-accuracy scores across audited containers
• Compliance checks run continuously rather than as a one-off project

Challenge

iOS 14+ privacy changes left ad platforms blind to a meaningful share of conversions, so Meta and Google campaigns were optimizing against an incomplete picture of what was actually working.

Approach

We built a server-side bridge that sends enriched, deduplicated conversion events directly to Meta CAPI and Google Ads, respecting consent state and hashing personal data before it ever leaves the server.

Results

• Recovers a meaningful share of previously lost iOS conversion signal
• Event Match Quality scores improved meaningfully after deployment
• One-click deployment for Shopify and WooCommerce storefronts, no developer required