NLP Solution For Pharmaceutical Marketing

Technologies used

Python

PostgreSQL

JavaScript

NodeJs

React

AWS

Motivation

The goal of the project was to create an MVP that uses AI and Machine Learning technologies to build assessment reports for pharmaceutical companies. The reports should be based on questionnaires and insights from doctors, helping the companies enhance trust in their products and improve marketing strategies.

Main challenges

Challenge 01

From Static Surveys to Intelligent Conversations

Traditional checkbox-based surveys were limiting the depth and authenticity of healthcare professional feedback. Respondents were forced into predefined answer options, leaving critical nuances, hesitations, and real-world clinical opinions uncaptured. This made it difficult for pharma brands to base strategic decisions on research that truly reflected HCP perspectives.

#1

Challenge 02

Scaling Insight Generation from Audio Responses

As the volume of open-ended audio responses grew, manual analysis became a bottleneck — slow, costly, and inconsistent across studies. Without an automated way to transcribe, cluster, and score hundreds of spoken answers, extracting actionable insights required significant analyst effort and introduced delays in the research cycle.

#2

Challenge 03

Exceeding the Limits of Commercial Research Tools

Existing commercial survey platforms were built for form-based research and could not support the combination of audio collection, AI-driven follow-up logic, and quantitative insight scoring the client needed. Relying on third-party tools meant accepting critical gaps in functionality, limited data ownership, and no path to the proprietary research product the client envisioned.

#3

The client's requirements

The client, launching a pharma market research, needed a technology partner to build the entire platform from scratch, not just code, but architecture decisions, AI strategy, and product thinking together. Specifically:

• Design and launch an online survey platform oriented around audio responses, not checkbox forms
• Integrate an AI agent capable of asking intelligent follow-up questions based on the completeness and depth of each respondent's answer
• Build a post-processing portal to cluster, score, and summarize qualitative responses at scale
• Generate presentation-ready reports (XLS/PPT) directly from collected data without manual analyst work
• Ensure the system could capture the authentic voice of HCPs rather than nudging them toward predefined answers
• Evolve from MVP to a mature product with sustained engineering and support

Why Azati?

Won a competitive tender through domain fit and solution quality

Azati was selected after winning a formal competitive process. The proposal stood out not just technically, but because the team demonstrated a credible understanding of pharma market research methodology and proposed a concrete, non-generic AI architecture.

End-to-end AI product engineering

The project required more than integration, it needed custom scoring models, clustering algorithms, and LLM-based conversational logic built from first principles. Azati brought ML engineers and AI specialists capable of owning that depth.

Long-term partnership mindset

Rather than delivering a fixed scope and stepping back, Azati embedded as a continuous development partner from MVP in 2021 through ongoing support, adapting the product as the client's research methodology matured.

Academic collaboration model

Azati structured a unique collaboration involving two PhDs in ML and NLP as intellectual advisors, while the engineering team acted as the implementation force. This ensured algorithms were rigorous and academically grounded, not improvised.

Building a research platform that truly listens?

Ready to move beyond checkbox surveys and capture the real voice of your audience with AI?

Let's design your AI research platform

Solution

01

AI-Driven Data Analysis

The system leverages LLMs and NLP to automatically process open-ended audio responses, clustering semantically similar answers, detecting key topics, and surfacing hidden insights at scale. Custom scoring algorithms prioritize findings by strategic relevance, eliminating manual analyst work. Models are continuously retrained based on evolving client research methodologies, ensuring output quality improves with every study cycle.

Key capabilities:

• Cluster responses semantically using LLM-based grouping for output efficiency
• Automatically extract key phrases and insights from doctor responses
• Assign relevance scores to insights for prioritization
• Continuously retrain models to align with updated research requirements

02

Conversational AI Survey Agent

An AI agent embedded in the survey flow evaluates each respondent's audio answer in real time and decides whether to ask an intelligent follow-up question. If a response lacks depth or leaves a strategic question unanswered, the agent probes further, mimicking the behavior of a skilled interviewer rather than a static form. This dynamic logic is powered by Claude LLM and LangChain, ensuring follow-ups are contextually relevant to each individual response.

Key capabilities:

• Evaluate response completeness and depth in real time
• Generate contextual follow-up questions dynamically per respondent
• Replace rigid branching logic with LLM-driven conversational flow
• Capture richer, more authentic HCP perspectives than checkbox surveys allow

03

Speech-to-Text Integration

Handles audio responses from medical practitioners using Whisper ASR, converting speech to text while recognizing domain-specific pharmaceutical terminology. This ensures that insights from spoken interviews are captured accurately and fed into downstream NLP pipelines without manual transcription.

Key capabilities:

• Accurately transcribe spoken responses with Whisper ASR
• Boost recognition for pharmaceutical and medical terms
• Normalize punctuation, capitalization, and sentence structure
• Seamlessly integrate transcribed data into NLP pipelines

04

Sentiment and Thematic Analysis

Applies NLP algorithms to evaluate sentiment and categorize clustered responses into strategic themes. This allows pharmaceutical companies to quickly identify positive, negative, or neutral feedback and understand which aspects of a product resonate most with healthcare professionals.

Key capabilities:

• Perform sentiment scoring on each response
• Group insights by product attributes, safety, and usage
• Highlight trends and anomalies in feedback
• Support marketing decisions with thematic analysis

05

Custom Report Generation

Generates structured XLS and PPT reports directly from clustered, scored data — no manual slide creation required. Reports are tailored for non-technical stakeholders in pharma marketing, presenting findings in a format that's immediately ready for strategic decision-making.

Key capabilities:

• Auto-generate XLS and PPT reports from processed data
• Rank insights by relevance and sentiment score
• Provide visual summaries and textual explanations
• Compare findings across respondent cohorts or product lines

06

Data Visualization and Sharing

Provides an intuitive portal for stakeholders to explore insights through dashboards. Users can filter, sort, and share visualizations, enabling faster decision-making and a clearer understanding of research findings across teams.

Key capabilities:

• Visualize key insights in interactive dashboards
• Share reports with stakeholders securely
• Highlight trends and key feedback points
• Support strategic planning and marketing optimization

Security

The platform is secured with SSL/TLS encryption and deployed on AWS infrastructure with access controls managed through AWS Guard. Environment monitoring is handled via CloudWatch, with a load balancer ensuring availability under variable respondent load. All audio recordings and sensitive HCP response data are stored in Amazon S3 with access policies aligned to the sensitivity of pharmaceutical market research data.

Engagement & delivery

Hybrid engagement model

Fixed Price for the MVP phase, T&M as the product matured, giving the client budget certainty early on and flexibility for long-term iteration.

Agile, continuous delivery

Iterative sprints with close client collaboration throughout:

• Platform, AI agent, and data portal developed in parallel
• Regular increments delivered for client testing and feedback
• Cross-browser QA via BrowserStack across devices and browsers
• Ongoing support alongside active development

Results & business impact

Automated Insight Generation

Reduced manual effort in analyzing medical questionnaires and automatically extracted actionable insights.

Enhanced Marketing Strategies

Insights from doctors allowed pharmaceutical companies to adjust campaigns and improve product positioning.

Improved Trust

Structured reports increased consumer confidence by showing alignment between medical recommendations and marketing claims.

Faster Decision-Making

Automated processing and visualization accelerated strategic decision-making for marketing teams.

Domain-Specific Accuracy

Integration of domain-specific NLP and speech-to-text ensured precise interpretation of medical feedback.

Strategic wins

The choice of incremental modernization over a complete rewrite ensured:

From questionnaires to conversations

Replacing checkbox surveys with open audio responses processed by an AI agent fundamentally changed data quality, HCPs could express nuanced opinions, making insights more authentic and strategically valuable.

Proprietary IP from day one

Custom scoring models and clustering algorithms mean the client owns a defensible technical asset, one that isn't replicable by switching providers.

Academic rigor embedded in the product

Two PhD advisors in ML and NLP ensured the analytical methodology met research-grade standards, critical in pharma, where insight credibility directly drives strategic decisions.

Startup to product company

From a founder with an idea to a fully operational SaaS, survey design, AI data collection, post-processing portal, and automated report generation, built and evolved over four years.

Team composition

A cross-functional team, backed by two PhD advisors, took the platform from zero to a fully operational AI-powered pharma research product over four years of continuous delivery.

• BA / Project Manager served as the client's main point of contact, gathered requirements, coordinated sprints, and ensured alignment between business goals and delivery.
• Backend Developer built the core platform logic and API layer using Node.js, powering survey design, data collection, and third-party integrations.
• Frontend Developer implemented the survey designer UI and respondent-facing interface in React, ensuring a seamless experience across devices.
• QA Engineer covered functional, regression, and cross-browser testing via BrowserStack across a wide range of devices, OS versions, and browsers.
• AI / ML Developer integrated Claude LLM and LangChain to power the conversational AI agent responsible for dynamic follow-up question logic.
• ML Engineer designed and implemented custom clustering algorithms, response scoring models, and the NLP pipeline for post-processing and insight extraction.
• DevOps Engineer managed AWS infrastructure, CI/CD pipelines, CloudWatch monitoring, and load balancing to ensure platform stability and scalability.
• Academic Advisors (×2), two PhDs in ML and NLP, provided research-grade guidance on algorithm design and validated the analytical methodology.