FinSight AI

Overview

FinSight AI is an agenticagenticDescribes AI systems that plan and take multi-step actions — calling tools, reading the results, deciding what to do next — rather than producing a single reply., multi-model financial intelligence platform that automates analysis which traditionally takes analysts hours. Enter a ticker and a date — get a probabilistic revenue forecast, a buy/sell signal, peer benchmarks, and a full audit trail in under 20 seconds. It combines quantitative finance, multi-agentmulti-agentA system in which several LLM-driven agents, each with its own role or tools, collaborate on a task and pass work between each other. LLMLLMLarge language model — a neural network trained on large amounts of text to predict the next token, used for generation, summarisation, classification and more. reasoning, and real-time market data into one fast pipeline.

The Problem

Traditional financial analysis means reading earnings transcripts by hand, building Excel models, monitoring news, and running competitive comparisons — repetitive work that's slow and hard to keep consistent. FinSight AI integrates all of it into a single automated pipeline with explainable outputs.

What I Built

• Parallel AI agents running concurrently via Python asyncio for sub-20-second results
• Probabilistic forecasts — Bear / Base / Bull scenarios via quantile regression, not a single point estimate
• SHAPSHAPSHapley Additive exPlanations — attributes a model's prediction to its input features using Shapley values from cooperative game theory. explainability so every forecast can be traced to its drivers
• News + NLP + competitor analysis for macro and peer context
• PDF reports with a full audit trail of how each signal was produced

Architecture

Specialized agents run concurrently and a final ensembler fuses their outputs into one calibrated signal:

• Transcript NLP agent — extracts and summarizes key financial drivers from earnings calls
• Financial model agent — quantitative forecasting with quantile regression (Bear/Base/Bull)
• News macro agent — aggregates, classifies, and scores macro news sentiment
• Competitor agent — benchmarks the target against industry peers
• Ensembler (CIO agent) — fuses all outputs into one calibrated investment signal with reasoning

Tech Stack

Frontend: React, Vite, Tailwind CSS

Backend: FastAPI, Python, asyncio

ML: Scikit-learn, quantile regression, SHAP

LLM: Groq (LLaMA 3)

Database: MongoDB Atlas

Use Case

Retail and analyst users who want a fast, explainable second opinion on a stock — forecast, signal, peer context, and the reasoning behind it — without spending hours building models by hand.

Status

Production (v2.0) with CI; actively maintained.