Transforming Search: A Journey to AI-Powered Conversational Experiences

Amrit kumar
Aug 1
4 min read

Updated: Aug 22

Just think! How humans express their thought?

In enterprise systems like contract management tools, search is often treated as a mechanical act. You find the bar, type in your query, filter endlessly, and hope for gold. But real users don’t think in keywords. They think in questions.

That’s what led me to rethink search altogether—not as a feature, but as a conversation.

Current Experience: Linear, Laborious, Limiting

Search flows today feel like solving a puzzle with missing pieces. Here’s how it typically goes:

Pre Search: Users must recall and locate the right search bar. Often, they're unsure of where to start and how to phrase their intent.
Entering the Query: They rely on system-specific keywords, metadata, or filters. This creates a knowledge barrier, especially for new users.
Search Result & Filter: Results come in bulk. Filters are rigid. Context is missing. Users must apply and reapply conditions just to get closer to what they really want.
Insight & Action: Once they find results, the data is raw. It lacks the storytelling or clarity needed for confident decision-making.

The video below explains the journey with existing screens

My UX Approach: Making Search Feel Like Thought

When I examined user behavior, I noticed something subtle: people weren’t really searching. They were voicing half-formed ideas — almost thinking out loud. Traditional filters and keyword search didn’t match that mental process.

So I reframed the problem:What if users could just ask — and the system understood the intent, retained the context, and served the answer?

That’s why I called it Thought. Unlike search, which matches keywords, Thought mirrors cognition:

Captures messy intent instead of requiring polished queries.
Retains context, like memory in a real conversation.
Responds visually and naturally, so answers feel immediate and human.

This shaped the foundation of AI-powered conversational search — where users type naturally (“Show me all Vodafone contracts with COLA clause”), the system understands nuance, remembers context, and responds in ways that reduce back-and-forth.

But here’s where Thought goes beyond tools like ChatGPT:

Domain-specific intelligence → It’s grounded in contracts and clauses, so answers are precise and trustworthy.
Visual + conversational hybrid → Results aren’t just text; they become clause explorers, dashboards, and summaries that lawyers can act on.
Task-oriented memory → It remembers context across workflows (“the NDA I asked about earlier”), not just within a single chat.
Explainability → Every answer shows why (the exact clause, rule, or data source), building trust in high-stakes legal work.

In short, Thought isn’t just about delivering answers — it’s about delivering trusted, actionable insights in the way professionals actually think and work.

The Change

Users now type naturally (“Show me all Vodafone contracts with COLA clause”).
The system understands nuance and remembers context.
The interface responds visually and contextually, reducing back-and-forth.

Landing page AI convo search — AI Search landing page

UX Enhancements That Shaped the Experience

To translate this vision into practice, I designed key UX enhancements:

1. Reworking the Mental Model of Search

Shifted from rigid, field-by-field filtering to natural, question-driven input.
Supported fuzzy, incomplete, and contextual prompts, so half-formed ideas still worked.
Built fallback UX for vague queries — instead of “no results,” the system suggested refinements.
Made chat context visible, so users always understood where they were in the flow.

2. Bringing Clarity to Data and Insights

Created response pages that combined summaries, filters, and metadata cards — like a structured search landing page.
Added inline interpretation panels so users could see what the system “understood” from their query.
Used color-coded visualizations to highlight risk patterns, clause distributions, and anomalies at a glance.

3. Designing for Transparency and Trust

Paired every AI answer with an explanation of logic (e.g., “COLA found under description X”).
Allowed users to inspect and adjust underlying filters to refine results with confidence.
Ensured the system was never a black box — trust was built into the design.

AI Search with a chain of reasoning for better trust and transparency.

The Future of Search: From Queries to Cognition

Search is no longer about typing keywords into a box. With AI, it can mirror the way people actually think — messy, contextual, evolving. The future lies in experiences where asking a question feels like continuing a train of thought, and the system answers with clarity instead of complexity.

The Role of AI in Enhancing User Experience

AI here is not just an engine; it’s the interpreter of nuance. It holds context across turns, adapts to the way users phrase things, and surfaces results in the form best suited to the task — a chart, a contract list, or a single clause. This transforms search into a thinking partner rather than a static tool.

Building Trust Through Transparency

In contract-heavy workflows, trust is everything. Users need to know why a certain contract surfaced or why a clause was flagged. By making AI’s reasoning visible — surfacing filters applied, highlighting keywords matched, showing “why this result” — the interface shifts from a black box to a transparent advisor. This reduces skepticism and builds confidence.

Continuous Improvement Through Feedback Loops

Great search isn’t “one and done.” Each query teaches the system — whether users refine, rephrase, or reject results. By designing lightweight feedback loops (like quick approval or disapproval signals), we allow users to shape the system without breaking flow. Over time, this creates a virtuous cycle: the more it’s used, the smarter and more aligned it becomes.

Design Details

Response page design decision

List of contracts on left as search response.

What I Learned (personal reflection)

Designing this project taught me that building AI-first experiences isn’t just about smarter algorithms — it’s about aligning with how humans actually think. I learned to design for messy intent, not polished input, and to make transparency as important as accuracy. Most importantly, I realized that trust in AI comes not from automation alone, but from explainability and control woven into the UX.