FinSight
A copilot trading workspace designed in three weeks for Inter IIT Tech Meet 13.0, with an AI that shows its reasoning, dockable gadget panels and visual automation. It won Gold.
UI/UX
SaaS
Role
Design (UX, frontend, PPT)
Timeline
Dec 2024 · 3 weeks
team
Inter IIT Tech Meet 13.0 team
platform
Web
What this is for
Professional traders and analysts work across many tools at once. They use trading platforms such as MetaTrader 5, Zerodha, TradingView and Bloomberg, along with news, broker feeds, spreadsheets and their own notes. Every decision means collecting data by hand, checking it, and then acting quickly. AI chatbots can answer questions, but they rarely show where an answer came from, which is a problem when money is at stake.
FinSight is a copilot trading platform: “A SaaS workspace for traders where a multi-agent RAG core ingests market data, news, broker/exchange feeds, user notes, and strategy docs to deliver explainable insights, automate routine tasks, and support safe, fast decision-making.” RAG (retrieval-augmented generation) means the AI looks up real sources before it answers.
Process: three weeks, three sprints
Brief. Turn a backend problem statement into something traders would actually use.
Sprint 1, discover. Secondary research covered trading platforms like MetaTrader 5, Zerodha, TradingView and Bloomberg, plus two or three research papers a day on AI in trading. Primary research meant interviews with 12 finance professionals: financial analysts, traders, AI researchers and a quant trader. These led to interpretations and a protopersona.
Sprint 2, define. Affinity mapping, user goals and stakeholder alignment, and a scenario produced two personas: Ranganath Shetty, a senior trader in Mumbai, and Shraddha Sinha, an analyst and part-time trader. Three themes stood out: fragmented workflows and personalisation, lack of trust in AI, and stress from repetitive work and decision-making.
Sprint 3, design. Solution development, wireframes, visual identity and animation in After Effects, prototyping and redesign. The interface went from V1 (a chat app) to V2 (a dense trading terminal) to V3.
Showing what the AI is doing
Trust was the biggest gap, so FinSight shows its work. While a query runs, a visible query pipeline traces each step: asking a clarifying question, breaking the question down, retrieving sources and grading documents, with an estimated time left. If something fails, a red dot marks it, and hovering explains what went wrong. Users can still choose Generate anyway, switch to a research mode for detailed prompts, and pick from suggested follow-up questions.
When an answer is ready to act on, users can move from chat mode to trade mode in one step. Interactive data gadgets also appear right inside the chat. These decisions draw on Nielsen’s heuristics: visibility of system status, error prevention, and user control and freedom.
A workspace that adapts to each trader
“Different traders need different set of tools,” so the workspace is built from gadget panels that users can add, dock, detach and rearrange, following Jakob’s law and the freedom to customise. Keyboard shortcuts and a right-click menu let users change chart types, export, annotate, or send a gadget to the AI to analyse and explain. The trade view lets traders draw patterns and indicators on charts and compare strategies in a sandbox.
Repetitive work moves into an automation studio. It’s a node-based canvas with templates such as DCF models and golden-crossover strategies, custom code nodes, and a prompt box for describing a model or strategy in plain words. The studio offers three ways to build an automation. Hick’s law and the Gestalt law of similarity shaped its layout.
The challenge
FinSight began as an Inter IIT Tech Meet 13.0 problem statement that only asked for a multi-agent RAG backend for financial queries. The team decided that wasn’t enough: “We will not just make the backend but a true Copilot for a trader. One that understands, explains, and acts in their workflow.”
Research showed how big the gap was. Of the 12 finance professionals interviewed, 91.67% said they don’t trust AI recommendations without clear backing data, and 91.67% had missed a trading opportunity that automation could have caught. 66.67% delay decisions because of information overload or uncertainty, and 66.67% spend time checking AI insights against live data by hand. The refined question: how might we automate and connect trading and analysis workflows so professionals can act faster, stay precise and focus on insights, instead of losing time to fragmented tools and repetitive manual tasks?
Outcome
FinSight won Gold at Inter IIT Tech Meet 13.0 in December 2024. The team built a working FinSight AI app that shows its backend status as the AI works, running on a RAG architecture made by the team.
Saptarshi’s part, as labelled on the team slide, was Design (UX, frontend, PPT). He worked alongside the team’s AI engineers and backend developers, with Shreyas as PS lead. The case study has no usability-test results or product metrics, so the Gold win is the only outcome claimed here.
Note: some market panels in the mockups were taken from a Figma Community file.
