Your documents, finally answering back

Upload documents, ask complex questions, and receive instant, citation-backed answers with real-time SSE token streaming and FlashRank reranking.

Architected with Industry-Leading AI Frameworks

Google Gemini
Pinecone Hybrid RAG
FlashRank Reranker
LangGraph StateGraph
Supabase Auth & RLS
Interactive RAG Workflow

How AskDoc Answers Your Questions

Experience an enterprise-grade retrieval pipeline engineered for speed, accuracy, and absolute precision.

1. Upload Documents

Server Validation

Drag & drop PDF, DOCX, XLSX, PPTX, or CSV files

Files are processed server-side with strict 10MB validation, split into semantic overlapping chunks, and vectorized instantly.

2. Hybrid Rerank & Quality Check

Rerank Loop

Pinecone Sparse/Dense Fusion + FlashRank Reranker

LangGraph queries Pinecone vectors & BM25 keyword indexes simultaneously, then re-scores top candidates using FlashRank before checking relevance quality.

3. Citation-Backed Response

SSE Streaming

Real-Time SSE Token Streaming + Citations

Gemini streams verified answers token-by-token alongside exact page citations and response metrics.

Pipeline Simulation — Step 1 of 3
Q3_Financial_Summary.pdf84 Chunks

Status: Parsed & Vectorized

Semantic Overlapping Chunks500 Tokens / 50 Token Overlap
PDF Table ExtractionOCR Scanned Fallback1024-dim Vectorization

Under the Hood Execution:

> Validating 10MB limit... Chunking 500 tokens / 50 overlap... Upserting Pinecone vectors.

Apple & Stripe Style Stacked Scroll

Architected for Uncompromising Quality

Scroll down to explore how AskDoc processes, reranks, and streams citation-backed answers.

01

Multi-Format Document Support

Ingest PDF, DOCX, XLSX, PPTX & CSV

Multi-Format Ingestion

Ingest any document type effortlessly. Server-side 10MB file validation occurs instantly before parsing text into 500-token semantic chunks with 50-token overlapping boundaries.

PDF / DOCX / XLSX
PPTX / CSV
10MB Server Validation
02

Pinecone Hybrid Vector Search

Dense Vectors + Sparse BM25 Keywords

Sparse + Dense RRF

Executes simultaneous dense embedding vector similarity queries alongside sparse BM25 keyword matching using Reciprocal Rank Fusion (RRF) for 99.4% precision.

Pinecone Indexing
BM25 Sparse Match
Reciprocal Rank Fusion
03

FlashRank Candidate Reranking

Ultra-Fast Context Re-scoring

FlashRank Engine

Reranks top candidate passages before passing context to the Gemini LLM. Eliminates irrelevant noise, preserving token budget and driving millisecond latency.

Context Re-scoring
Noise Elimination
<1.2s Total Latency
04

LangGraph StateGraph & Checkpointer

Stateful Reasoning & Ambiguity Detection

Postgres Checkpointer

Orchestrates intent routing, quality-check loops, ambiguity detection, and persistent Postgres checkpointer thread memory for multi-turn conversations.

Intent Classification
Ambiguity Check
Multi-Turn Memory
Live Product Preview

See Real-Time Streaming & Citations

Try sample prompts below to test SSE token streaming, FlashRank citations, and ambiguity detection in action.

AskDoc Live Session
Model: Gemini 3.6 Flash
What is the Q3 revenue and growth rate?
Q3_Financial_Report.pdf (Page 12)
0.85s 142 tokens $0.000042
Engineering Architecture

Under the Hood Pipeline

A multi-stage LangGraph StateGraph workflow built for latency, precision, and state persistence.

User Query

FastAPI Route

Intent Classifier

LangGraph Node

Pinecone + BM25

Reciprocal Rank Fusion

FlashRank Reranker

Top Candidate Scoring

Gemini LLM

SSE Streaming

Ready to Unlock Your Documents?

Try AskDoc on Your Own Documents Now

Upload your PDF, DOCX, XLSX, PPTX, or CSV files and start asking questions in seconds with zero setup.

User-scoped Supabase Row-Level Security Enabled