CASE STUDY
WINGMANAI
Industry
Professional Networking · AI
Location
Kolkata, India
Development Time
Full-Stack Build
Cooperation Period
2024
ABOUT THE PROJECT
WingManAI is a production-ready AI platform that captures professional conversations with mutual consent, transcribes them with speaker diarisation, and delivers 19-dimension analysis spanning language, acoustics, and relational dynamics. Growth Loops Technology designed and built both backend services — a NestJS application server and a Python AI microservice — from initial architecture to production deployment. The platform addresses a universal gap in professional networking: meetings happen, but almost nothing actionable comes out of them. WingManAI captures the full richness of a real conversation, transcribes it, analyses it acoustically and semantically, and surfaces structured insights that help professionals improve how they communicate and connect.
TECH STACK
Client Vision
The client envisioned a platform that transforms how professionals reflect on and learn from their conversations. Despite being a cornerstone of professional growth, in-person conversations leave almost no structured record — professionals walk out of networking meetings with vague impressions and little to act on. The vision was to capture the full richness of real conversations, analyse them acoustically and semantically across 19 dimensions, and surface structured insights that help professionals measurably improve how they communicate and connect.
Our Execution
Growth Loops Technology architected and built WingManAI from the ground up across two purpose-built backend services. The NestJS application server handles all product logic — authentication, user management, meeting scheduling, real-time communication via WebSockets, and Stripe subscription billing across three tiers. The Python AI microservice runs the full processing pipeline: speaker-diarized transcription via AssemblyAI, word-level acoustic analysis using librosa, speaker recognition with SpeechBrain, and a two-tier LLM analysis system powered by LangChain and GPT-4o. A multi-stage BullMQ queue pipeline with retry logic ties the two services together, delivering real-time progress to the frontend while processing runs asynchronously in the background.
WingManAI is built for ambitious professionals who attend networking events, client meetings, and industry conferences — and want more than a business card to show for it. The platform is designed for anyone who wants to understand not just what was said in a conversation, but how they communicated: their speaking style, emotional tone, persuasion patterns, and conversational dynamics.
To transform every professional conversation into a structured, actionable record — capturing acoustic and linguistic dimensions that reveal how professionals communicate, not just what they say.
Professional meetings leave almost no structured record. Existing tools either record passively or analyse text after the fact — none connect the acoustic, linguistic, and relational dimensions of a conversation into a coherent, actionable picture.
Professionals need a consent-first platform that records conversations, transcribes with speaker identification, and delivers deep analysis of communication dynamics — all within a seamless, privacy-respecting experience.
Leverage AI and signal processing to give professionals a genuine competitive edge in how they communicate — turning every meeting into a coaching opportunity backed by data.
OVERVIEW
WingManAI transforms professional networking by capturing conversations with mutual consent, transcribing with speaker diarisation, and delivering 19-dimension analysis spanning language, acoustics, and relational dynamics — giving professionals structured, actionable insights from every meeting.
WingManAI's backend is composed of two purpose-built services that work in concert. The NestJS application server handles all product logic — authentication, user management, meeting scheduling, real-time WebSocket communication, and subscription billing — while the Python AI microservice handles all GPU-intensive, long-running AI workloads. Keeping them decoupled allows each to scale independently.
Each conversation is analysed in two sequential LLM tiers powered by GPT-4o via LangChain with structured Pydantic output. Tier 1 covers speaking style, key themes, language patterns, and interaction dynamics. Tier 2 goes deeper — persuasion strategies, conflict areas, agreements reached, and future communication steps. Both tiers run per-user and conversation-level analysis in parallel using asyncio, covering 19 distinct dimensions in total.
A word-level acoustic analysis pipeline built with librosa extracts pitch, volume, and RMS energy per frame, mapping these values onto each sentence and word from the diarized transcript. Per-speaker statistics are normalized to z-scores for cross-conversation comparability. SpeechBrain's speaker embedding model fingerprints each user's voice, mapping AssemblyAI's generic labels to real user identities using cosine similarity.
PRODUCT GLIMPSE
WingManAI: Turning professional conversations into structured intelligence — capturing, transcribing, and analysing every meeting across 19 dimensions of language, acoustics, and relational dynamics to give professionals a genuine edge in how they communicate.
Analysis Dimensions
Pipeline Reliability
Transcription Accuracy
Production Readiness
MEET OUR TEAM
Ayush Agarwal
Project Delivery Manager
Shivam Kumar
Backend Lead
Debjit Konar
AI/ML Developer
Arannyak Roy
Backend Developer
Murtuza Siddiquee
QA Engineer