Speed vs depth
A composite score for 3-second decisions. Drill-down for deep work.
04 / ELIROX LLC / 6 WEEKS
Elirox AIgo Terminal.
A 0→1 AI trading terminal that gives retail F&O traders institutional decision quality, with a risk engine the AI cannot override.
Lead UX DesignerProduct design, full-stack build, AI/ML integrationFigJam · AI/ML · Real-time systems
The impact
−0%Decision time per trade: 47 minutes down to 4.1.
The problem
89% of retail F&O traders lose money, juggling six or more fragmented tools with no synthesis layer.
The solution
One context-aware terminal: explainable signals, a persistent AI chat rail and a ten-check risk engine.
0%
faster decisions (47 → 4.1 min)
6 → 1apps while trading
78%check FII data daily
94%rate AI explanations useful
Headline results from the 8-week pilot · screens to follow
- My role
- I led UX and shaped the product end to end, from 24 contextual inquiry sessions to the system model, across design, build and AI/ML integration.
- The central decision
- The AI can be swapped. The risk engine cannot. Safety rules are designed in, not bolted on.
- Project scope
- A web application for NSE F&O markets, built across a 6-week engagement and validated in an 8-week pilot with 18 traders.
01 / Problem
Retail traders lose to tool inequality, not strategy.
Institutions have terminals and quant desks. Retail traders juggle six or more fragmented tools with no synthesis layer.
The gap
Pick a stat. 24 contextual inquiry sessions, Mumbai and Bangalore.
0 of 100 traders
of retail F&O traders lose money
SEBI study, 20230 min
per informed trade decision
0+
apps open at once
0 min
entry window missed, up to
“By the time I check NSE for FII data and flip to TradingView to confirm the breakout, the candle is already closed.”
Retail trader, contextual inquiry
02 / Strategy
Three tensions. Three answers.
How might we give retail traders institutional decision quality at 1/100th of the cost?
Automation vs control
Every AI signal shows its factors and weights.
Beginner vs expert
Dense panels for experts. Natural language for novices.
03 / Design
The AI can be swapped. The risk engine cannot.
Five layers feed one decision surface. Safety rules were designed in from day one, not added later.
System model
Each layer has one job.
Data
NSE ticks · options chain · FII/DII feed
Signal
Technical · FII pressure · options flow · news NLP
AI
Explains signals and drafts trade plans
SwappableRisk
SEBI limits · kill switch · Greeks caps · margin buffer
FixedInterface
Data-dense dashboard · AI chat rail · live ticks
0%
more content area than top tabs
0×
faster than polling
0
pre-trade checks per order
0%
daily loss cap, kill switch
Kill-switch simulator
Move the daily loss past the 4% cap. The AI still says BUY, and the risk engine still blocks the order. Illustrative trade.
AI engine · swappable
BUY SBIN
Risk engine · fixed
- Kill switch · 4% daily cap
- Market hours
- Concentration
- Margin buffer
- Lot caps
- + 5 more checkpoints
“The kill switch blocked me once. I was furious in the moment. Two weeks later I realised it saved me.”
Pilot trader, week 6
04 / Impact
An 8-week pilot with 18 traders.
Measured against a diary-study baseline taken before adoption.
Run the decision
One informed trade decision, start to finish. Scaled so 1 second = 10 minutes.
Before · six apps
ZerodhaNSETradingViewSensibullMoneyControlExcel
0.0 minAfter · one terminal
AIgo Terminal
0.0 minBefore and after
Each measure is scaled to its own range.
BeforeAfter
Decision time per trade
Apps used while trading
Daily FII data check
Pre-trade risk checks per trade
View as table
| Measure | Before | After |
|---|---|---|
| Decision time per trade | 47 min | 4.1 min |
| Apps used while trading | 6 | 1 |
| Daily FII data check | 17% | 78% |
| Pre-trade risk checks per trade | 0.3 | 1 |
Success metrics
Baseline, target and result.
| Metric | Baseline | Target | Result | Delta |
|---|---|---|---|---|
| Decision time per trade | 47 min | <10 min | 4.1 min | −91% |
| Daily FII data adoption | 17% | 60% | 78% | +61 pp |
| Risk checks before order | 0.3 / trade | 1.0 / trade | 1.0 / trade | +233% |
| Signal accuracy (backtested) | — | 55%+ | 62% | +7 pp vs target |
| API latency (p95) | — | <200 ms | 148 ms | 26% under target |
| Kill-switch false positives | — | <5% | 2.1% | Under target |
| AI response accuracy (user-rated) | — | 80%+ | 94% | +14 pp vs target |
05 / Learnings
What worked. What we got wrong.
Three lessons carried into the v3 roadmap.
Worked: context injection
Every prompt carries live market context. That is the AI moat.
Worked: transparency
Showing factor weights built trust faster than accuracy.
Wrong: desktop-first
14 of 18 traders checked on mobile. Dense grids break below 768 px.
06 / PROCESS
How it came together.
- 01
Weeks 1–2
Data pipeline and FII/DII pressure engine.
- 02
Weeks 3–4
Options flow and signal fusion.
- 03
Weeks 5–6
AI strategy engine and 10-checkpoint risk layer.
- 04
Weeks 7–8
Backend and dashboard v1: 8 modules, AI chat.
Contact