Intelligence Engineered
Into Every Workflow.
20+ AI/ML systems deployed in production — from BERT-powered NLP pipelines and LLM-native SaaS products to Agentforce agents that make real business decisions, autonomously.
AI that Scales to Production
From NLP pipelines handling millions of queries to LLM-powered SaaS products — real AI implementations delivered across enterprise workflows.
AI-Powered JEE Admissions Counsellor with RAG-Grounded LLMs
admissionmantrana.ai (CareerMantrana)
Challenge
Over 1 million JEE aspirants every year face a paralysing decision funnel: 100+ NITs, IIITs, and GFTIs, six rounds of JoSAA counselling, float-vs-freeze calls, and category-state-rank permutations that existing rank-predictor sites flatten into static forms. Anxious students and risk-averse parents were left with generic college lists, no contextual guidance, and no clear next step — while high-intent leads evaporated before ever reaching a human counsellor.
Solution
Built admissionmantrana.ai for CareerMantrana: a conversational AI admissions counsellor that replaces form-based rank predictors with natural-language Q&A grounded on JoSAA cutoff data via a retrieval-augmented LLM pipeline. Built with a conversion-optimised growth funnel — JoSAA round countdown urgency, risk-surfacing suggested prompts, in-answer CTAs on borderline results, and a credit-wall that routes anxious queries into paid counselling bookings.
Implemented for
CareerMantrana
OTT Customer Support Intelligence with BERT NLP
Hotstar, Sony LIV, MX Player
Challenge
Millions of customer queries flooding support channels during live events — manual triaging caused severe delays, inconsistent classification, and agents spending 70%+ of their time on spam and repetitive tickets rather than real issues.
Solution
Deployed a Google BERT-powered NLP pipeline integrated with Salesforce and Freshdesk: automated spam filtering, multi-level query classification, semantic intent mapping, and intelligent routing — with supervised + semi-supervised models that continuously improve from live data.
Implemented for
LLM-Powered Salesforce Org Documentation Engine
TechnicalDebts.com
Challenge
Salesforce orgs accumulating years of undocumented Apex classes, LWC components, and custom flows — leaving teams with high developer dependency, painful onboarding, and runaway technical debt that compounded with every new release.
Solution
Built TechnicalDebts.com: an AI-native SaaS product using Meta Llama 3.1 that auto-scans Salesforce org metadata, generates structured contextual documentation for every component, and continuously monitors changes — with a human + AI validation layer for accuracy.
Agentforce & Conversational AI Agents
Agentforce, GPT-4, Claude, Azure ML
Challenge
Enterprises sitting on valuable CRM data but lacking the AI layer to act on it — repetitive decisions made manually, support queues handled by humans, and AI tools deployed in silos without integration into real business workflows.
Solution
Custom-built Agentforce agents, AI copilots, and decision engines embedded directly into Salesforce workflows — autonomously handling support routing, lead qualification, case summarization, and operational decisions, with fallback escalation and full audit trails.
Implemented for
