🌿
আদর
DOCINTEL
Document Intelligence Platform
Tap anywhere to begin →
🌿
আদর
DOCINTEL
1 / 27
🌿
Welcome
আদর DocIntel
Your intelligent document analysis platform — upload, classify, detect language, embed, and chat with your documents using AI-powered semantic search
500
Docs per user
pgvector
Vector search
5
Languages
Step 1
Secure Registration & Login
Create your private workspace in seconds. Every account is isolated — your documents are never shared with other users.
🔐JWT-based authentication — sessions persist across page refreshes
🛡bcrypt password hashing — your credentials are never stored in plain text
👤Role-based access — regular users and system administrators
docintel.adar.agomoniai.com
🌿
আদর
DocIntel
Email
user@example.com
Password
••••••••••
Sign in →
No account? Create one
Step 2
Upload, Redact, Classify & Detect Language
Drag and drop files or click to browse. Every file is stored securely in Google Cloud Storage, then optional PII redaction can protect extracted text before chunking, embedding, classification, and AI workflows.
📄Supports PDF, DOCX, CSV, images, and plain text
🔒Redact PII before chunking and embedding — emails, phones, SSNs, cards, DOB, bank IDs, IPs
🌐Detects English, Spanish, Bengali, Hindi, Arabic, and other ISO language codes
🧭Labels invoices, leases, resumes, reports, policies, and more
☁GCS path: users/{uid}/documents/{did}/source/
docintel.adar.agomoniai.com
আদরDocIntel
📂 Documents💬 Chat
3
Total
1
Chunked
2
Embedded
⬆ Upload
⬆
Drop files or click to upload
PDF · DOCX · CSV · Images · TXT · 497 slots remaining
🔒 Redact PII before chunking
Lease_Extension_931_Bryant.pdf
Embedded ✓Lease · LegalEnglish · 42 chunks · PII redacted
Step 3
Automatic Chunking with Metadata
Immediately after upload, DocIntel extracts text, applies optional PII redaction, and splits protected content into 350-word semantic windows with 60-word overlap — no manual step needed.
✂Smart word-window chunking preserves sentence boundaries
🔒Spacing-aware redaction catches OCR variants like jane @ example . com and 123 45 6789
🏷Each chunk carries full metadata: doc ID, user ID, filename, file type, position
💾Chunks saved to GCS: chunks/chunk_0001.txt
GCS: chunks/_metadata.json
// _metadata.json
"document": {
  "id": "91f2d246-9d40...",
  "user_id": "a4b8c2d1...",
  "filename": "Q3_Report.pdf",
  "file_type": "pdf",
  "total_chunks": 42
},
"chunks": [
  { "index": 0, "word_count": 347,
    "gcs_path": "...chunks/chunk_0000.txt" },
  { "index": 1, "word_count": 351,
    "gcs_path": "...chunks/chunk_0001.txt" },
  ... 40 more chunks
]
Step 4
Browse Source & Chunks
Open the original document via a signed GCS URL, or browse every individual chunk with word count, storage path, and an on-demand PII privacy toggle for safe review.
🔗Source button opens a signed URL valid for 1 hour — no direct cloud access needed
📋Chunks panel shows all 42 chunks — click any to view full text content
🔒Chunk viewer PII toggle masks sensitive identifiers while inspecting raw text
☑Select multiple chunks for targeted chunk-level summarization
Chunks Viewer
42 Chunks
#1 · 347w
#2 · 351w
#3 · 348w
#4 · 352w
#5 · 344w
Chunk #1 of 42
🔒 PII on
GCS: users/uid/documents/did/chunks/chunk_0000.txt
Contact [REDACTED_EMAIL] · Phone [REDACTED_PHONE] · SSN [REDACTED_SSN]

The Q3 2024 financial results demonstrated strong performance across all business segments. Revenue grew 17% year-over-year to $2.8 million, driven primarily by expansion in enterprise accounts and increased adoption of premium tier subscriptions...
Step 5
One-Click Vector Embedding
Click Embed on any chunked document to generate vector embeddings. These are stored in pgvector — PostgreSQL's vector extension — enabling semantic search.
⚡Powered by Gemini embedding-2 — 768 dimensions via REST API
🐘pgvector with HNSW index for cosine similarity search
🔒Every vector is user-scoped — you can only search your own documents
Embedding pipeline
📄
Annual_Policy_2024.docx
18 chunks · Ready to embed
⚡ Embed
Embedding progress
Chunk 12 of 18 — storing in pgvector...
✓ Chunk 0000 → [0.023, -0.441, 0.182, ...768 dims]
✓ Chunk 0001 → [0.187, 0.063, -0.294, ...]
→ Chunk 0012 → embedding...
Step 6
Voice + Multilingual Chat
Ask by typing or speaking in English, Spanish, Bengali, Hindi, Arabic, or the language of the document. A chat privacy toggle redacts PII from questions, history, retrieved context, source previews, and traces before model use.
🎙Spanish mic input — Web Speech where supported, MediaRecorder + Gemini transcription fallback for Safari and Firefox
🔒PII toggle masks structured identifiers, including spaced/OCR-style formats
🔍pgvector cosine similarity — finds the most relevant chunks to your question
🌐Response language follows the selected documents when appropriate
📎Every answer cites its source chunks with similarity percentage
💬 Chat
✓ Q3_Financial × 🔒 PII Annual_Policy
■ Recording Spanish voice… transcribes with Gemini fallback
¿Cuál fue el ingreso de Q3 y su crecimiento?
🌿
Los ingresos de Q3 2024 fueron $2.8M, un crecimiento interanual de 17%, impulsado por expansión enterprise y adopción del plan premium [Source 1].
🐘 pgvector · 3 sources · 94%
Feature
Rich Table Rendering
When the AI detects tabular data — financials, schedules, comparisons — it automatically formats the response as a proper readable table.
📊Markdown tables rendered as clean HTML tables with alternating rows
✍Also renders bold, italic, bullet lists, headings, and inline code
🎯System prompt instructs the model to prefer tables for structured data
Chat — table response
Show revenue breakdown by quarter
🌿
Here is the quarterly revenue breakdown [Source 1]:
QuarterRevenueGrowthMargin
Q1 2024$2.4M+12%24%
Q2 2024$2.8M+17%26%
Q3 2024$3.1M+11%28%
Operating margins improved steadily through cost optimisation [Source 1]
Feature
Multilingual Document Summarization
Summarize any chunked document — no embedding required. Choose from five types, write a custom prompt, and turn on PII privacy before sending content to Gemini.
⚡Executive — 3-5 sentence overview with key conclusions
📑By Section — identifies every section and summarizes each
🌐English, Spanish, Bengali, Hindi, and Arabic summaries with exact citations preserved
🔒Summary privacy mode redacts document text and custom prompts before generation
✏Custom — write your own: "What are the financial risks?"
📝 Summary Panel
Summarize: Q3_Financial_Report.pdf
⚡ Executive • Key Points 📑 By Section 📄 Detailed 🔒 PII ✏ Custom
▶ Generate Summary
EXECUTIVE SUMMARY✓ Complete
আদর DocIntel Q3 2024 demonstrated strong financial performance with revenue of $2.8M, up 17% year-over-year. Operating margins expanded to 28% driven by enterprise growth and cost optimisation. Management raised full-year guidance citing strong pipeline visibility.
Admin
System Administration Dashboard
Administrators get visibility across users, documents, usage, audit history, and AI traces — including question-level inspection of retrieval, prompts, tool calls, and LLM responses.
📊Live stats — total users, documents, embedded vectors, storage used
👥Promote or demote users, delete accounts with full data cascade
🗑Delete any document — removes GCS files, chunks, and pgvector rows
🧭Traces tab — select a question and inspect spans, retrieved context, tool calls, and LLM output
⚙ Admin Dashboard
24
Users
1,247
Docs
891
Embedded
84K
Vectors
UserRoleDocsActions
admin@adar.comadmin34—
user@example.comuser12↑ Promote
dev@company.comuser87🗑 Delete
Feature
Secure Password Reset
Forgot your password? No problem. আদর DocIntel sends a secure reset link directly to your email — valid for one hour, single use only.
📧One-click forgot password from the login screen — enter your email and a reset link is sent instantly
🔒Tokens are SHA-256 hashed in the database — the raw token only ever exists in the email
⏱Links expire in 1 hour and are single-use — clicking a used link shows an invalid token error
🌿Sent via Google SMTP from your domain — fully branded আদর DocIntel email
Password Reset Flow
① Login screen
Password Forgot password?
••••••••
Click "Forgot password?" to begin reset
② Email sent
From: আদর DocIntel <admin@agomoniai.com>
Subject: Reset your আদর DocIntel password
Reset Password →
Link valid for 1 hour · Single use · SHA-256 hashed
③ Set new password
New password (min 8 chars)
Set new password →
Security
Built-in Rate Limiting
Every endpoint is protected against abuse. Brute-force login attempts, credential stuffing, and API flooding are automatically blocked per user or IP.
🔐Login: 10 attempts per minute per IP — blocks brute-force credential attacks
📝Register: 3 per minute — prevents mass account creation
💬Chat: 30 queries per minute per user — fair usage across all users
📂Upload: 20 files per minute — prevents storage abuse
⚡Returns HTTP 429 Too Many Requests with a clear retry message
Rate limit response
Rate limits — per endpoint
Endpoint Limit Key
POST /auth/login10/minIP
POST /auth/register3/minIP
POST /auth/forgot-password5/minIP
POST /chat/stream30/minUser
POST /documents/upload20/minUser
POST /summarize/*/stream15/minUser
HTTP 429 — Too Many Requests
{"detail": "Too many requests — max 10 per 60s. Please wait."}
Feature
Persistent Chat History
Every conversation is saved to the database and restored across devices, browsers, and sessions. Your chat history is always there when you need it.
💾Conversations auto-saved to PostgreSQL 1.5 seconds after each AI response
📱Access the same history from any device — not localStorage, real server storage
✏Sessions are auto-titled from your first question — easy to find later
🗂Up to 50 sessions listed per user, ordered by most recent — click any to restore
🗑Delete any session — removes all messages from the database permanently
💬 Chat — Session History
History ＋ New
What was Q3 revenue?
4 msgs · May 20
Show policy sections
6 msgs · May 19
List all deadlines
2 msgs · May 18
Key financial risks
8 msgs · May 17
✓ Q3_Financial ×
What was Q3 revenue?
🌿
Q3 revenue was $2.8M, a 17% increase YoY [Source 1].
✓ Saved · 4 messages
Ask a question…
➤
Retrieval
Hybrid Search: Vector + BM25 + RRF
Pure cosine similarity misses exact matches — product codes, invoice numbers, names, dates. আদর DocIntel fuses semantic vector search with PostgreSQL full-text search using Reciprocal Rank Fusion.
🔍pgvector cosine similarity finds semantically similar chunks — great for concepts and meaning
🔤PostgreSQL BM25-style full-text search finds exact keyword matches — product codes, names, dates
⚡Reciprocal Rank Fusion (RRF) combines both lists: score = 1/(60+vector_rank) + 1/(60+fts_rank)
🏆Chunks found by both methods score highest — hybrid badge shows when both signals agree
↩Automatic fallback to vector-only when FTS returns zero results (stopword-only queries)
Hybrid Search Pipeline
Query: "Invoice #INV-2847 payment terms"
pgvector
Cosine Similarity
top 50 semantic matches
+
PostgreSQL GIN
BM25 Full-Text
top 50 keyword matches
▼ Reciprocal Rank Fusion ▼
Re-ranked unified top 20
score = 1/(60+v_rank) + 1/(60+fts_rank)
Source badges in chat
⚡ hybrid Invoice_2847.pdf 94%
Payment terms for invoice #INV-2847: Net 30 days from receipt...
🔤 keyword Contract_Terms.pdf 61%
All invoices must reference INV prefix followed by numeric ID...
🔍 vector Payment_Policy.pdf 57%
Payments are due within the agreed net terms upon delivery...
Quality
Cross-Encoder Re-ranking via Gemini
Retrieval gets you candidates. Re-ranking gets you the right answer. A cross-encoder reads the question and each chunk together — it sees context that bi-encoders miss.
🎯Bi-encoders score query and chunk independently — fast but context-blind
🧠Gemini cross-encoder reads (query + chunk) together — scores true relevance 0.0 to 1.0
📊Fetch 20 candidates from hybrid search, Gemini re-orders → top 6 sent to LLM
✦Re-ranked badge appears on source cards — shows which chunks survived the re-ranking stage
↩Graceful fallback to retrieval order if Gemini re-rank fails
3-Stage RAG Pipeline
①
Hybrid Retrieval
pgvector + BM25 + RRF → 20 candidates
↓
②
Gemini Re-ranking
scores each (query, chunk) pair → top 6
gemini-1.5-flash
↓
③
LLM Generation
top 6 re-ranked chunks → grounded answer
After re-ranking
✦ re-ranked ⚡ hybrid Lease_2025.pdf 91%
✦ re-ranked 🔍 vector Addendum_A.pdf 74%
Architecture Positioning
Horizontal RAG Platform + Vertical Agentic AI
DocIntel keeps the reusable self-service RAG foundation separate from specialized domain workflows. Healthcare users can record multilingual visits, generate clinical-scribe packets, ask cited questions across records and payer guides, and run prior authorization agents from the same grounded workspace.
↔Horizontal RAG — upload, OCR, chunking, embeddings, hybrid search, rerank, 5-language voice/chat, summaries, comparison, citations
🔒Horizontal controls — PII redaction, auth, workspaces, usage limits, traces, audit logs, observability, governance
🎙Clinical scribe — record or upload visit audio in English, Spanish, Bangla, Hindi, or Arabic, then create transcript, SOAP draft, patient summary, and follow-up checklist
🏥Clinical agents — patient context, clinical summary, labs, medications, follow-ups, care gaps, administrative and governance flags
🧾Prior authorization agents — payer criteria, evidence mapping, missing documentation, submission risks, and human-review packet
💬Chat asks: does this meet prior auth criteria, what evidence supports it, what documentation is missing, and what next action is needed?
⚙Workflow — JSON orchestrator runs subagents, saves structured outputs, evaluates quality, and supports human approval
👥Beneficiaries — patients, caregivers, clinicians, small clinics, prior auth teams, billing/admin teams, and compliance reviewers
Layered Intelligence Architecture
Lease Management
Abstracts
Clause review
Dates + obligations
Healthcare
Clinical scribe
Patient story
Prior auth
Brokerage
KYC
Disclosures
Suitability review
Horizontal RAG + Governance Platform
Upload OCR PII Embeddings Hybrid RAG Rerank Citations 5-language voice Traces Audit Billing
Pattern
Purpose
Control
JSON config
Orchestrator, subagents, tools, output contract per vertical
Reusable
Domain agents
Clinical scribe, clinical packet, and prior auth packet with saved state, approval, and evaluation
Agentic
Guided chat
RAG retrieves cited visit transcripts, patient records, and payer evidence, then uses saved workflow context
Grounded
Approval
Human review saves clinical findings, payer evidence, gaps, risks, and packets
Governed
Evaluation
Backend scores completion, citations, confidence, safety, governance, and approval readiness
Audited
Quality Loop
Feedback on AI Responses
Every AI response includes a thumbs-up and thumbs-down button. Your ratings create a quality signal that identifies bad retrievals and poor answers — building a continuous improvement loop.
👍Thumbs up or down on every AI message — click once to rate, click again to change
💾Ratings saved to PostgreSQL with the triggering question and answer for context
🔄Persistent across sessions — your ratings are restored when you reload or switch sessions
🎯Selecting one rating dims the other — clear visual state: green for helpful, red for not helpful
📊Admins can review all feedback via the API — thumbs counts, recent ratings, and user emails
💬 Chat — Feedback
🌿
Q3 revenue was $2.8 million, a 17% increase year-over-year, driven primarily by enterprise contract growth in the APAC region.
✓ Marked helpful
🐘 pgvector 2 sources ▾
🌿
I couldn't find specific information about the Q3 cost breakdown in the provided documents.
✗ Marked not helpful
Saved to PostgreSQL
message_id: "abc123xyz"
rating: +1 (helpful)
question: "What was Q3 revenue?"
restored: ✓ on next session load
Export
Export Chat & Summaries
Every conversation and summary can be exported as a Markdown file. Take your AI-generated insights into Notion, Confluence, Word, or any editor — with sources included.
↓Export any chat session as Markdown — click the Export button in the chat top bar
📄Includes all questions, answers, and source citations (document name, chunk, relevance %)
📝Export any summary as Markdown — button appears in the summary panel after generation
⚡Instant client-side generation — no server round-trip, no waiting
🌐Open .md files in Notion, Obsidian, VS Code, Typora, or paste into any editor
Chat Export — Q3_Revenue_Analysis.md
Chat top bar
✓ Q3_Financial.pdf 3/3
↓ Export
# Q3 Revenue Analysis
*Exported from আদর DocIntel · May 20, 2026*

---

**You:** What was Q3 revenue?

**আদর DocIntel:** Q3 revenue was
$2.8M, up 17% YoY...

> **Sources:** Q3_Report.pdf
> (chunk 3/12, 94%)

---
Summary panel header
Executive Summary — Q3_Financial.pdf
↓ Export .md
✕
Metering
Strict Usage Metering & Analytics
Every costly action is tracked and enforced in real time. Users click 📊 Usage to see documents, file size, chat queries, embedding chunks, summaries, compares, lease AI, voice transcriptions, evals, storage, and all-time activity.
📊Usage panel in the header — always one click away, loads live stats from PostgreSQL
📤Tracks documents, upload size, embedding chunks, chat, summaries, compares, lease AI, voice, and eval cases
🧱Quota is reserved atomically before expensive model work starts, preventing concurrent bypass
🌡Progress meters turn amber at 80% of your plan limit — visual warning before hitting the wall
🛑Over-limit requests return clear 403, 413, or 429 errors before consuming AI capacity
📊 Usage & Limits
Your plan: Pro Plan
📂
12
Documents
of 500
💬
23
Queries today
of 500
Enforced plan limits
Documents 12 / 500
Queries today 23 / 500
Embedding chunks today 84 / 100
8/50
Compares
4/50
Lease AI
16/250
Voice
All-time activity
📤 Uploads47
⚡ Embeddings1,284
💬 Queries312
⇄ Comparisons8
🏢 Lease AI17
📊 Eval cases24
Billing
Tiered Limits & Plan Management
Three plans — Free, Pro, and Enterprise — with limits enforced per user in the database. Admins upgrade any account instantly, and JSONB overrides can tune individual quotas without redeploying.
🆓Free: 20 docs, 10 MB files, 50 queries/day, 10 embedding chunks/day, 5 summaries/day, 3 compares/day
⚡Pro: 500 docs, 500 MB files, 500 queries/day, 100 embedding chunks/day, 50 summaries/day, 50 compares/day
🏢Enterprise: unlimited docs, queries, summaries, compares, lease AI, voice, and evals; 10 GB files
⚙Admins change any user's tier with a dropdown in the Users table — takes effect immediately
🔧Lease AI, voice transcription, and eval quotas are enforced alongside core RAG usage
🛑Exceeding limits returns HTTP 403, 413, or 429 with a clear upgrade message
⚙ Admin → Users
FREE
20 docs
10 MB
50 q/day
5 summaries
POPULAR
PRO
500 docs
500 MB
500 q/day
50 summaries
ENTERPRISE
∞ docs
10 GB
∞ q/day
∞ workflows
Email
Role
Tier
Docs
admin@agomoniai.com
admin
12
user@company.com
user
3
HTTP 403 — Document limit reached (20 docs on Free tier). Upgrade to upload more.
HTTP 429 — Daily query limit reached (50 on Free tier). Try again tomorrow.
Atomic check — quota is reserved before model calls, embeddings, voice transcription, evals, or lease agents run.
Collaboration
Team Workspaces
Create shared workspaces where your team collaborates on the same document set. Each workspace has role-based access — owners manage membership, editors upload and embed, viewers can only read and chat.
🏢Create a workspace, invite teammates by email — they must have an existing account
👑Owner — full control: rename, delete, manage members, upload, embed, and chat
✏Editor — can upload documents and run embeddings, but cannot manage members
👁Viewer — read-only: upload button hidden, can chat and search shared documents
🔀Switch context with one click — workspace badge appears in the header, chat history and documents scope to that workspace automatically
🗑Delete protection — must type the workspace name exactly to confirm; warns about other members losing access
🏢 Workspaces — Q3 Finance Team
Workspaces
🏠 Personal
🏢
Q3 Finance
3 members · 8 docs
🏢
Legal Review
2 members · 3 docs
New workspace
Team name…
＋
Q3 Finance Team
3 members · 8 documents · your role: owner
✏ Rename
✓ Active
🗑 Delete
Member
Role
Change
admin@agomoniai.com
owner
—
analyst@firm.com
editor
editor ▾
✕
partner@client.co
viewer
viewer ▾
✕
Invite member
colleague@company.com
viewer ▾
+ Invite
👑 Owner
Full control
✏ Editor
Upload + embed
👁 Viewer
Read + chat
Trust
Email Verification
Every new account must verify their email address before logging in. A 24-hour link is sent automatically on registration. Admins are auto-verified. Existing accounts are grandfathered in.
📧Verification email sent immediately on registration — powered by Gmail SMTP
🔒Unverified accounts receive HTTP 403 on login with a clear message to check inbox
🔄Resend button available — new 24-hour link sent without revealing if email exists
👑First user (admin) is auto-verified — no chicken-and-egg on fresh deployments
🛡Tokens stored as SHA-256 hashes — raw token never persisted to the database
📧 Check your inbox
📧
Verify your email
We sent a link to
analyst@firm.com
Click the link to activate your account.
It expires in 24 hours.
Back to login
Email received
Verify your আদর DocIntel email address
From: আদর DocIntel <admin@agomoniai.com>
Welcome to আদর DocIntel!

Please verify your email address by clicking the link below:

https://docintel.adar.agomoniai.com/verify-email?token=abc123...

This link expires in 24 hours.
HTTP 403 — Please verify your email address before logging in. Check your inbox or use the resend link.
Ops & Compliance
Privacy, Audit & Traces
Audit logs explain what users did. PII redaction protects sensitive identifiers. Traces explain how an AI answer was produced, including whether privacy mode was enabled for a question or summary.
🔔In-app toast when embedding completes: "Q3_Report.pdf is ready for chat (47 chunks)"
📧Email notification sent to uploader via Gmail SMTP when status changes to embedded
📬Workspace invite email sent when a member is added to a team workspace
🔍Audit log table in Admin dashboard — action, user, resource, IP address, timestamp
🔒PII redaction — upload-time and runtime privacy modes for chunks, chat, summaries, comparisons, and traces
🧭Trace tab — filter by question, open a trace, and inspect spans, retrieved chunks, LLM/tool events, prompts, and response
🧩How it works — X-Trace-Id connects trace_flows, trace_spans, and trace_llm_events for one chat or voice request
🗄Stored in PostgreSQL audit_log and trace tables — query directly for compliance and debugging exports
⚙ Admin → Traces
✅
Embedding complete
"Q3_Financial.pdf" is ready for chat (47 chunks embedded)
Question Traces
Filter by question
↻ Refresh
Question
Flow
Status
Trace
Q3 revenue growth?
RAG
success
trc_8fa…
Spanish contract summary
voice
traced
trc_2bc…
Invoice payment terms?
retrieve
review
trc_91d…
Why traceability matters: prove the answer came from the right evidence and identify the exact failure point
X-Trace-Id → trace_flows → trace_spans → trace_llm_events → Admin question view
Organisation
Document Recovery & Tags
Errored documents can be retried with one click — the original file is re-downloaded from GCS and re-processed. Tags let you organise hundreds of documents into named collections with colour-coded chips, filtering, and sorting.
↻Retry button on error-state documents — wipes stale chunks, re-downloads from GCS, re-runs the full chunk → embed pipeline
🏷Create colour-coded tags — assign multiple tags to any document with a single dropdown click
✕Remove tags from documents instantly — the ✕ on each chip unassigns without deleting the tag
🔍Live search bar filters documents as you type — no page reload
⇅Sort by date, name A–Z, file size, or status — combined with tag filter for precision
🗄Tags stored in PostgreSQL — survive sessions, shared across devices, visible per-user
📂 Documents — Tags & Filter
🔍 Search…
🏷 Q3
↓ Date
🏷3
Q3 Finance (3) ✕ Legal (5) ✕ HR (2) ✕
New tag name…
＋ Add
📄
Q3_Financial_Report_2026.pdf
Embedded ✓ 🏠 Personal PDF · 2.4 MB · 47 chunks
Q3 Finance ✕ Legal ✕
🔗 Source
📋 Chunks
📝 Summary
↩ Re-embed
📝
corrupted_report.docx
Error Chunking timeout after 30s
↻ Retry
🔗 Source
🗑 Delete
Monetisation
Stripe Billing — 3 Plans
Full Stripe Checkout integration with three tiers, a 3-day free trial on paid plans, and automatic plan activation on login — backed by the same strict quota enforcement shown in the Usage panel.
🆓Free — 20 docs · 10 MB files · 50 queries/day · 5 summaries/day · 3 compares/day
🚀Pro ($20/mo) — 500 docs · 500 MB files · 500 queries/day · 50 summaries/day · 50 compares/day
🏢Enterprise ($100/mo) — unlimited docs, queries, summaries, compares, lease AI, voice, and evals; 10 GB files
🎁3-day free trial on Pro & Enterprise — card collected, no charge until day 4
🔄Auto-sync on login — Stripe subscription status checked on every sign-in, tier updated instantly
🔒Cancel anytime via Stripe portal — downgrade to Free reflected on next login automatically
💳 Plans & Billing
Free $0
  • 20 documents · 10 MB files
  • 50 queries · 5 summaries · 3 compares/day
✓ Active
Most popular
Pro
$20/mo
  • 500 documents · 500 MB files
  • 500 queries · 50 summaries · 50 compares/day
🎁 3-day free trial — no charge until day 4
Start 3-day free trial →
Enterprise
$100/mo
  • Unlimited docs · 10 GB files
  • Unlimited queries, workflows, voice, evals
🎁 3-day free trial — no charge until day 4
Activation Flow
Stripe Checkout → Redirect to login → Plan active on sign-in ✓
Quality Assurance
Evaluation: Answers + Agent Workflows
DocIntel evaluates both conversational RAG answers and vertical agent workflow runs. Inline answer evals help users judge response quality; backend-persisted workflow evals support governance, approval gates, traceability, and audit history.
🎯Relevance — does the answer actually address the question asked?
🔍Specificity — does it give concrete details, numbers, and dates vs vague generalities?
⚖️Confidence — is uncertainty expressed proportionally, neither overconfident nor over-hedged?
🧩Coherence — is the summary or comparison logically structured with no contradictions?
📊On-demand Evaluate button in summary and comparison panels — fires only after streaming completes
🏆Graded 1–5: Excellent · Good · Acceptable · Poor · Fail — hover for judge reasoning
🧭Agent workflow eval API — stores overall score, gate status, metric scores, recommendations, and policy threshold
💬 Chat — with inline eval scores
What was the revenue growth in Q3 compared to Q2?
📚
Based on the Q3 Financial Report, revenue grew 18.4% from Q2 to Q3, reaching $4.2M. Operating margin improved by 2.1 percentage points to 24.3%.
Eval
🎯 5/5
🔍 5/5 ⚖️ 4/5 ↻ ✕
🎯 Relevance 5/5
Excellent ✓

Directly and completely addresses the revenue growth question with specific figures

Does the answer address the question?

After summarization completes:
📊 Evaluate
→ 🧩 4/5 🔍 5/5
Agent workflow evaluation
91%
Overall
Pass
Gate
82%
Citations
Saved
Audit
POST /api/agent-evals/lease/runs/{run_id} persists metrics, recommendations, and policy.
🌿
আদর DocIntel
Everything you need to work smarter with documents
A complete self-service Document Intelligence Platform — from upload to insight in minutes.
📂
Upload & Store
  • Up to 500 documents
  • PDF, DOCX, CSV, Images
  • Secure Google Cloud Storage
  • Signed URL access
✂
Chunk & Index
  • Auto 350-word windows
  • Rich metadata per chunk
  • GCS path storage
  • Browse all chunks
⚡
Embed & Search
  • Gemini Embedding-2
  • 768-dim vectors
  • pgvector + HNSW index
  • Cosine similarity search
🎙
Voice Chat
  • Speak + auto-submit
  • Safari/Firefox fallback
  • Gemini transcription
  • Source-cited answers
📝
Summarize
  • 5 summary types
  • Executive to Custom
  • Map-reduce for large docs
  • No embedding needed
⚙
Admin Control
  • Real-time system stats
  • User management
  • Document oversight
  • Role-based access
🔐
Security + PII
  • JWT + bcrypt auth
  • Upload-time masking
  • Chat + trace privacy
  • OCR spacing aware
🗂
Chat Sessions
  • PostgreSQL-persisted history
  • Auto-titled from question
  • Access from any device
  • Up to 50 sessions per user
🔀
Hybrid Search
  • pgvector cosine similarity
  • PostgreSQL BM25 full-text
  • Reciprocal Rank Fusion
  • ⚡ hybrid / 🔤 keyword / 🔍 vector badges
✦
Gemini Re-ranking
  • 20 candidates → Gemini scores
  • Cross-encoder (query + chunk)
  • Top 6 sent to LLM
  • Fallback to retrieval order
⚙
Vertical Agents
  • Lease + healthcare agents
  • Brokerage ready
  • Evaluation gates
  • Human approval + save
👍
Feedback Loop
  • Thumbs up/down on AI responses
  • Persists across sessions
  • Saved to PostgreSQL with context
  • Admin feedback analytics API
↓
Export
  • Chat → Markdown with sources
  • Summary → Markdown .md
  • Client-side, instant download
  • Notion / Obsidian / VS Code ready
📊
Usage Metering
  • Tracks uploads, queries, embeds
  • Today vs all-time breakdown
  • Storage bytes per user
  • Live usage panel in header
🏷
Tiered Plans
  • Free / Pro / Enterprise tiers
  • Per-user DB enforcement
  • Admin tier change dropdown
  • HTTP 403/429 on limit breach
🏢
Team Workspaces
  • Owner / Editor / Viewer roles
  • Invite teammates by email
  • Workspace-scoped docs + chat
  • Switch context in one click
📧
Email Verification
  • Verify before first login
  • 24-hour token via Gmail SMTP
  • Resend button available
  • Admin auto-verified
🧭
Traceability
  • Question-level traces
  • Spans + LLM events
  • Retrieved chunk inspection
  • Admin trace viewer
💳
Stripe Billing
  • Free / Pro $20 / Enterprise $100
  • 3-day free trial on paid plans
  • Auto-sync tier on every login
  • Cancel → auto-downgrade to Free
📊
Inline Eval
  • Auto-scores every chat answer
  • 🎯 Relevance · 🔍 Specificity · ⚖️ Confidence
  • Grades 1–5 with judge reasoning
  • On-demand for summaries & compare
↻
Doc Recovery
  • Retry button on error docs
  • Re-downloads from GCS
  • Re-runs full chunk → embed
  • No delete + re-upload needed
🏷
Document Tags
  • Colour-coded tag collections
  • Assign/remove per document
  • Search · filter · sort bar
  • Stored in PostgreSQL per user
🚀 Try আদর DocIntel free 🌿 docintel.adar.agomoniai.com
🐘 pgvector☁ Google Cloud🤖 Gemini AI⚡ FastAPI⚛ React🔥 Firebase