SYED_ABBAS
← All work
CASE STUDY · [01]
YEAR · 2024
ROLE · Full-stack + RAG

CX Consulting AI

Full-stack CX consulting platform with RAG, multi-LLM routing, and real-time document processing.

Outcomes
0%Retrieval accuracy
0msp95 latency
0+Concurrent users
0%Uptime
MEDIA SLOT
cx-consulting hero (screenshot or 1080p demo video)
Drop a wide screenshot or product demo into /public/projects/cx-consulting-ai/hero.jpg
BRIEF

Consultants needed an AI assistant they could trust to cite the right paragraph from long policy documents. The bar was reliability, not novelty.

[01]

The product

Consultants upload documents, ask questions, and get answers with citations they can verify against the source. The retrieval pipeline was tuned against a client evaluation set until top-1 accuracy hit 91%. Multi-LLM routing picks the model per request based on the task. A newly uploaded policy PDF is queryable within a few minutes of ingest.

MEDIA SLOT
chat / citation UI
Suggested: main chat interface with a citation open
MEDIA SLOT
document upload flow
Suggested: 2-3 frames from ingest to queryable
[02]

Getting to 620ms

The first working version answered in about 2.3 seconds at p95, which was too slow. Getting to 620ms took reworking the retrieval path, caching hot embeddings, and moving simpler requests onto a smaller model.

p95 response latency
milliseconds, lower is better
First working build
2,300
After caching pass
1,400
After model routing
950
Current
620ms
STACK
FastAPILangChainChromaDBNext.jsPostgresRedis