Finlet AI: An Evidence-Driven Multi-Agent Financial Intelligence Platform Using Model Context Protocol and Deterministic Risk Governance (International Journal of Advanced Research in Science Communication and Technology)

Summary

• Finlet AI is a twelve-layer multi-agent platform that separates market arithmetic and primary-source retrieval from neural text synthesis.p.1
• The risk equation weights normalized volatility (45% weight), news sentiment risk (35% weight) and regulatory filing flags (20% weight).p.3
• A composite risk score of 0.65 or more routes the workflow to a human approval gate before any report is produced.p.5
• Live execution on Reliance Industries (RELIANCE.NS) showed a price of ₹1,184.50 (−2.03%) and 30-day volatility of 23.1%.p.7
• Steady large-cap equities produced composite risk scores between 0.268 and 0.376 (LOW) and were auto-approved.p.10
• NVDA reached a score of 0.670 (HIGH) and was halted at the SQLite human-in-the-loop checkpoint.p.10

Key insights

1. Ungrounded conversational models quote outdated prices, invent page or footnote numbers and give inconsistent risk ratings on identical inputs.p.1
2. Three domain workers (price series, news feeds, annual reports) run in parallel behind standardized Model Context Protocol tool interfaces.p.1
3. The deterministic risk formula applies weights of 45%, 35% and 20% and emits an inline factor audit trail.p.3
4. Finlet AI is the only system in the comparison table with a deterministic risk mechanism and a stateful human-in-the-loop gate.p.4
5. Scores of 0.65 or higher serialize the full graph state to SQLite and return a PENDING status until an analyst approves or rejects.p.5
6. Document pages are indexed with 384-d all-MiniLM-L6-v2 embeddings so that filing citations keep their exact page.p.6
7. Scores from 0.40 are classified MEDIUM and scores from 0.65 are classified HIGH.p.8
8. The design is intended to guarantee 100% score repeatability across runs.p.7