Sony MAX
Filmykaant Chatbot
Sony MAX, one of India's most popular Hindi movie TV channels, wanted a chatbot on Facebook Messenger that could engage Bollywood fans at scale — answering film trivia, sharing live TV schedules, running quizzes, and building a community of engaged viewers. The project required solving a language problem that no vendor had solved before: reliable NLP for Hinglish.
Quick facts
- Client
- Sony MAX
- Role
- Project Manager
- Category
- Conversational AI
- Year
- 2017
- Duration
- 6 months build + ongoing
2M+
Unique users
1M+
Daily conversations
Top 10
Facebook globally
FBStart
Programme selected
Background & context
In 2017, Facebook had recently opened the Messenger platform to bots, and brands were racing to build engaging conversational experiences. The challenge for Sony MAX was that their audience — primarily Hindi-speaking fans aged 18-35 — communicated entirely in Hinglish: Hindi written informally in English script. "Bhai yaar Amitabh ki best movie kaun si hai?" was a perfectly normal query. No NLP platform at the time — not Dialogflow (then API.ai), not Wit.ai, not IBM Watson — could reliably understand it.
The challenge
What made this hard.
01
The Hinglish NLP gap
Hinglish is not a formal language with standardised grammar or spelling. It spans a spectrum from pure phonetic transliteration ("aaj raat kya hai") to casual code-switching mid-sentence ("bhai this movie ka ending kya tha?"). No training dataset existed. No pre-trained model existed. We had to build the language understanding capability from scratch.
02
Designing for viral scale
Sony MAX's audience for major film events could spike to hundreds of thousands of concurrent users. The platform architecture needed to handle dramatic traffic spikes without degradation — and the conversation design needed to be engaging enough that users would actually tell their friends.
03
Content freshness and real-time integration
A Bollywood bot that gives yesterday's TV schedule or last week's quiz is worthless. We needed real-time integration with Sony MAX's broadcast feed, a content management system that could be updated without engineering involvement, and a quiz engine with fresh questions daily.
The approach
How we solved it.
Building the Hinglish NLP from the ground up
We ran structured data collection campaigns — asking Bollywood fans to type questions they would naturally ask a chatbot. We collected and labelled thousands of utterances across intent categories: film queries, actor queries, schedule requests, quiz participation, smalltalk, and a long tail of everything else. We built a custom intent classifier and entity extractor tuned specifically for Hinglish phonetic patterns, common abbreviations, and code-switching.
Three-tier fallback architecture
High-confidence intent match → direct response. Low-confidence match → clarifying question before responding. No match → graceful acknowledgement, menu redirect, and automatic logging for training data review. The third tier was a data collection mechanism, not a failure mode. Every unhandled query improved the next model version.
The Filmykaant persona
We defined a character: Filmykaant, the Ultimate Bollywood Fan. Opinionated, enthusiastic, slightly dramatic — exactly what the audience expected from a Bollywood entity. The persona gave the bot a consistent voice and made interactions feel like talking to a character rather than querying a database.
Gamification for viral growth
The daily Bollywood quiz with a public leaderboard was the growth engine. Users competed, shared their scores, and challenged friends. Streaks and achievement badges created daily return visits. The viral coefficient of the quiz feature drove organic user acquisition throughout the campaign.
Impact & outcomes
What we delivered.
2 million unique users at peak, 1 million conversations per day
India's highest-traffic entertainment chatbot at the time of launch. The scale was achieved through product design, not paid acquisition.
Facebook Top 10 Chatbot globally
Filmykaant was recognised by Facebook as one of the top 10 chatbots worldwide on the Messenger platform — a significant distinction at a time when thousands of bots were competing for attention.
Selected for Facebook FBStart programme
Facebook's support programme for high-potential Messenger applications. Selected based on user engagement metrics, technical architecture, and growth trajectory.
Custom Hinglish NLP became a reusable capability
The NLP model built for Filmykaant became the foundation for subsequent Hinglish-language projects at Cedex Technologies, creating a durable competitive advantage.
Tools & technologies
"The Filmykaant bot exceeded every expectation we had for engagement and scale. The team solved a language problem we thought was unsolvable, and built something our audience genuinely loved."
Sony MAX Digital Team
Sony Pictures Networks India
Lessons learned
What this taught me.
The best conversational products are not the ones with the most capable AI — they are the ones that know exactly what they are good at and design gracefully around the edges of that capability.
Gamification works because it taps into social proof and competition, not just habit formation. A leaderboard that your friends can see is more motivating than points only you can see.
Building training data collection into the product from day one — treating every unhandled query as a data point — creates a compounding improvement loop that rule-based systems cannot replicate.
Next case study
Enterprise GenAI Platform →