NeuralQuant
AI-Powered Stock Intelligence Through Adversarial Multi-Agent Debate
Institutional-grade quant engine + PARA-DEBATE multi-agent system + live voice analysts. US + India. 100% live data.
This codebase is available for acquisition. Contact the author for details.
Live at neuralquant.co · API neuralquant.onrender.com (v4.1.3)
Current Status — September 2026
Production live. API v4.1.3, ~950 stocks in the score cache (≈450 US + ≈500 India), nightly refresh healthy. Last full prod smoke (2026-08-18): 15/15 endpoints PASS, zero JS console errors across the app.
Recently shipped (Sessions 90–109):
- Branding unified to NeuralQuant; landing, methodology, pricing refreshed.
- Voice analysts live — QuantAstra (on-demand) + Veronica (page-aware companion with "Hey Veronica" wake word + morning briefing), via a LiveKit worker. TTS migrated from ElevenLabs to LiveKit Inference (cartesia/sonic-3.5).
- Hermes — autonomous self-improving trading agent surfaced at
/hermes (live "Matrix" dashboard, SSE log stream). Migrated from expired Railway trial to a GCP Always Free e2-micro VM.
- India data parity — QuantFactor universe expanded from ~60 to ~500 NSE names via a daily-refreshed sheet sync. Legacy
anjali_enrichment collapsed onto quantfactor_universe.
- Security hardening (P0–P6) — Supabase RLS, log redaction, gitleaks + dependency scanning in CI, IDOR fixes, HTTP security headers + CSP (report-only), per-IP abuse limiting, webhook-signature enforcement, a security-event audit log, and an incident-response runbook. See Security.
- Score-scale bug family fixed — composite_score normalization that broke trade Kelly sizing, meta scores, and backtest hit rate.
- GCP India price feeder — Render's outbound IPs are blocked by Yahoo, so a self-contained feeder on the GCP VM now fetches Indian prices and writes them to
stock_snapshot.
- Screener —
market=BOTH/GLOBAL merges US + IN; empty/500 bugs fixed.
- QA pass — fixed false "loss-making" badge, CSP console error, "Sign In while authenticated" nav bug, null
change_pct, broken SSE stream, empty stock_snapshot cold-start; removed the unused /alerts page and all email functionality.
- Mobile — hamburger nav + bottom tab bar + responsive cards/charts/tables across the app; tightened cramped score grids on phones (verified live at 390px).
Operator follow-ups (not code): manual-deploy quantastra-agent Render worker if not yet deployed, final deletion of any remaining Railway account. The GCP India price feeder is now functional. Before any live demo, run python scripts/warmup.py (~3–5 min prior) — first-hit Ask Morgan / PARA-DEBATE is ~50s / ~85s of inherent multi-agent LLM time, and warming the exact demo tickers makes the audience-facing clicks fast.
Acquisition Overview
AI-powered stock intelligence platform — India + US.
Multi-agent Claude-powered research engine (PARA-DEBATE) · QuantFactor quintile engine + proprietary IRS% score across ~950 stocks · web + mobile (Expo) + voice (LiveKit) + an autonomous trading agent on one FastAPI backend · Anthropic Claude with an AWS Bedrock path. See neuralquant.co/methodology for the research approach.
18+ months of build. 90+ documented sessions.
Evaluate in 15 minutes — zero API keys
git clone <repo> && cd NeuralQuant
cp .env.example .env # DEMO_MODE=true is the docker-compose default
docker compose up --build
# → http://localhost:3000
The demo stack bundles a real data snapshot (Supabase tables, ~1,669 rows) and a canned PARA-DEBATE output, so the dashboard, screener, stock detail, and backtest pages work offline. Add ANTHROPIC_API_KEY to run live agent debates.
What's included in the sale
Codebase · neuralquant.co domain · Supabase data + Q1FY27 backtest results · scoring methodology IP (IRS%, G Score, Risk Efficiency, QuantFactor quintiles) · sister repo (anjali-value-stocks) · brand/design system · optional 3–6 month founder transition. Full list: docs/ASSET_INVENTORY.md. Due diligence: docs/ARCHITECTURE.md · docs/BUG_HISTORY.md · docs/OPERATIONS.md · docs/HANDOVER.md. Patent strategy: IP_PROTECTION/MEETING_PREP.md.
What is NeuralQuant?
A full-stack AI stock intelligence platform that combines a quantitative signal engine with a multi-agent AI analyst debate (PARA-DEBATE), natural-language and voice analysts, and an autonomous trading agent — backed entirely by live, real data.
Why It's Different
| Feature | NeuralQuant | Danelfin | TipRanks | Seeking Alpha |
|---|
| Adversarial agent | Mandatory BEAR challenge | No | No | No |
| India market | ~500 NSE names, live | No | No | No |
| Natural language queries | Yes (Ask Morgan) | No | No | No |
| Voice analyst | Yes (QuantAstra + Veronica) | No | No | No |
| LLM hallucination guard | Per-agent metric reconciliation vs live data | No | No | No |
| Real-time enrichment | RSI/MACD/ATR live | End-of-day | End-of-day | End-of-day |
| Multi-agent debate | 6 agents + head analyst, 1 verdict | Single score | Aggregation | Single author |
| Pricing | Free / $9.99 / $29.99 / $99.99 | Free / $20 / $49 / $99 | Free / $30 / $50 | $239/yr |
This is the beginning of the README. Read the rest at the original. Read the full README ↗
