Built for
Industry: Healthcare & Medtech · Delivered: Multi-modal Medical-History Intake
Natural Language Repair Search Across 160K+ Records for Diesel Laptops
Diesel Repair Chat replaces dropdown-filter navigation with natural language chat. Type a symptom in plain English and the system extracts vehicle attributes, searches 160,000 repair records, and returns the matching procedure — inside your AWS VPC.
42+ PROVIDERS • 493 INTERVIEWS • HIPAA-COMPLIANT • PATENT-PENDING
The problem
Technicians Waste Time With Dropdown Filters Instead Of Repair Information.
Current workflow. Diesel Repair users rely on dropdown filters to find procedures based on vehicle attributes and symptoms. This process is inefficient, error-prone, and frustrating when navigating datasets of 160,000+ technical records.
The solution. Replace dropdown navigation with a natural language chat interface. Technicians describe the issue in their own words, and the system extracts the vehicle attributes and symptom, searches the dataset, and returns the matching procedure with source attribution.
Natural Language To Repair Procedures.
Diesel Repair Chat—a hybrid search pipeline running entirely in your AWS VPC. The system combines vector embeddings for semantic matching with structured metadata filtering to retrieve repair information from 160,000 records in under 2 seconds.
Attribute Extraction From Free-Text
LLM automatically extracts make, year, model, type, and symptom from natural language queries. Handles misspellings, abbreviations, and synonyms intelligently.
Hybrid Search Across 160K Records
Vector embeddings combined with structured metadata filters for semantic search. Returns ranked results with source attribution (row number and document title).
Conversational Session Management
Follow-up questions merge into the same session automatically. Review or clear history via the sessions endpoint.
Secure REST API
Every endpoint requires JWT authentication with role-based access for admin functions. Ask, manage sessions, check health, and submit feedback via the API.
Built-In Monitoring & Feedback
Every query and response is logged for analysis. Feedback and Prometheus-compatible metrics support continuous improvement.
Five Essential Search Platform Features.
Sub-2-second query target.
Optimized vector indexing on 160K records targets sub-2-second responses for 95% of queries.
Fuzzy matching through the LLM
The parser accommodates misspellings, abbreviations, and non-standard terminology, and asks a follow-up when a query is genuinely ambiguous.Conversational context across sessions
Session management preserves conversation history and merges new details into prior queries across multiple interactions.
Thumbs up/down feedback on every response.
Feedback is logged with its query and response for continuous improvement.Data sovereignty in your AWS VPC
All 160K records, embeddings, and queries stay inside Diesel Laptops' AWS environment. No data leaves the VPC without explicit authorization.
Ready To Ship Natural Language Search On Your Repair Data?
30-minute consultation. Bring your dataset, your AWS environment, and your use cases. Walk out with a working architecture and fixed-scope timeline.
Our process
Four Phases From Requirements To UAT In Your AWS Environment.
Data Ingestion
Architecture Design
Build & Integration
UAT & Deployment
Fixed-scope delivery. From CSV ingestion through vector indexing, FastAPI development, query pipeline integration, and AWS VPC deployment to UAT—12-week engagement.
What's inside
Production Stack & Tech Partners.
Python and FastAPI backend with Qdrant for vector storage and search. The query processing pipeline handles parsing and attribute extraction; AWS Bedrock handles response generation. Docker containerization and Prometheus/Grafana monitoring support production reliability and observability.
Figure 1. FastAPI query pipeline (NL parser, guardrails, attribute extraction) · session context management · Qdrant hybrid search (160K records) · AWS Bedrock response generation with guardrails · source attribution · all inside AWS VPC.
Results
Dropdowns To Conversational Search.
Before
Legacy Dropdown Workflow.
Technicians manually navigate dropdown filters searching 160K records. Inefficient, error-prone, and frustrating. Memorization required. Complex for rare queries.
AFTER
Natural Language Search.
Type any symptom to get ranked repair procedures in seconds. LLM extracts attributes intelligently. Results include source attribution. Session history preserved.
Latency
Target response time for 95% of queries
Records
Repair procedures indexed in Qdrant
Throughput
Target capacity under normal load
Concurrency
Concurrent users without degradation
Availability
Uptime target during business hours
Recovery
Target mean time to recovery from failures
Recognised By The Platforms That Vet AI Companies
Kodexo Labs is reviewed where technical buyers do their diligence: Clutch and Upwork. Every badge below links to the live profile.
Ship Natural Language Search On Your Repair Or Operational Data?
30-minute consultation. Walk through your dataset, AWS environment, and performance targets. Get a fixed-scope timeline and architecture blueprint.
Frequently asked
Sub-2-second target for 95% of queries. Vector similarity search combined with metadata filtering returns ranked results from 160K records in seconds, staying within 15% even as data volume grows 5x.