AutoAdvisor MVP

AUTOadvisor

Intent

Matches

Brand BOT

Deal

FREE Are you a car brand? List with us →

MVP · Step 1 Live

Step 1 of 4 — Intent Capture

What's your ideal car?

Describe what you're looking for in plain language. Our AI extracts your requirements automatically.

Tip: mention budget, use-case, fuel preference, seating needs

Finding your best matches…

Sending to gpt-4o-mini

Running NLU slot-filling

Validating UserProfile schema

Computing slot confidences

Extracted User Profile — feeds into Module 3 (Vector Search)

⚠ Some slots have low confidence. You can proceed or refine your input for better matches.

Data flow: Step 1 → Step 2

▼

INPUT — raw user text

"I want a safety car in 15L for my family with good mileage"

OUTPUT — UserProfile (Pydantic model → stored in session)

{ budget_max: 1500000, use_case: "family", vehicle_type: "suv",
seating_capacity: 5, fuel_preference: null,
priority: ["safety", "mileage"], slot_confidences: {
budget_max: 0.95, use_case: 0.91, priority: 0.88 } }

Step 2 of 4 — Recommendations

Finding your matches...

Finding the best cars for you…

Building profile embedding

Vector search in ChromaDB

Scoring + ranking matches

Generating match justifications

Data flow: Step 2 → Step 3

▼

INPUT — UserProfile passed to ChromaDB vector search

{ budget_max: ..., vehicle_type: "...", priority: [...] }

OUTPUT — BotContext (on card select → passes to Brand BOT)

Awaiting match results…

Step 3 of 4 — Brand Advisor

Brand Advisor

📸 Brand Media

Selected Car

Best Offer Discussed

—

Data flow: Step 3 → Step 4

▼

INPUT — Bot context sent to /chat API each turn

Select a car from Step 2 to see the live context.

OUTPUT — DealContext built from conversation (passes to Step 4)

Chat in progress — deal context updates here as negotiation proceeds.

WHERE DOES THE BOT GET ITS ANSWERS?

1. Car specs → fetched from ChromaDB by sku_id (ground_clearance, torque, airbags, mileage etc.)
2. Brand voice → persona config in api_server.py (tone, key talking points per brand)
3. User profile → budget, priorities, dynamic requirements from Step 1
4. Conversation history → all previous turns sent with each message (multi-turn memory)
5. LLM → gpt-4o-mini generates response grounded in the above — it CANNOT invent specs

Step 4 of 4 — Deal Summary & Decision

Confirm your details to lock this offer

Your dealer will contact you within 2 hours to finalise.

Preferred Test Ride Date

Deal Locked ✓

Lead created successfully

Data flow: Step 4 output → Lead System (leads.json)

▼

INPUT — DealContext A + B (from Step 3)

deal_a: {brand:"Hyundai", price:1464000, perks:[...]}
deal_b: {brand:"Kia", price:1450000, perks:[...]}
cot_winner: "CRETA_2024" confidence: 0.87
trade_off: "Creta costs ₹14k more but 0.5 kmpl better mileage"

OUTPUT — LeadPayload appended to leads.json

{ lead_tag: "HOT_LEAD", engagement_score: 90,
modules_completed: ["intent","recommend","brand_bot","deal"],
chosen_sku: "CRETA_2024", offered_price: 1464000,
user: {name:"...", phone:"...", city:"..."},
timestamp: "2024-01-15T14:32:00Z" }