MAILSENSE
Loading…
AI-Powered Gmail Intelligence

Draft perfect
email replies,
evaluated by AI.

MailSense reads your Gmail, generates context-aware replies using Mistral AI, then scores every draft through a deterministic evaluation suite before saving it as a ready-to-send Gmail draft.

80%+
Evaluation threshold
11
Quality metrics scored
Self-correction loops
HOW IT WORKS
From inbox to perfect replyin four automated steps.
01

Fetch

MailSense connects to Gmail via Composio OAuth and pulls your most recent unread threads automatically.

02

Retrieve

Pinecone vector search finds the top-3 most similar email pairs from your dataset to use as few-shot context.

03

Generate

Mistral AI drafts a professional reply using your active prompt template, few-shot examples, and tone policy.

04

Evaluate

11 metrics — BLEU, ROUGE, BERTScore, LLM Judge + 8 deterministic rules — score the draft. Fails trigger a self-correction loop.

EVALUATION ENGINE
11 metrics. Zeroguessing. Always.

Every reply is scored across lexical, semantic, and qualitative dimensions before it ever reaches your drafts folder.

Lexical Metrics

Pure token-overlap measures to catch word-level drift.

BLEU-4 n-gram precision
ROUGE-L longest common subsequence
METEOR stem alignment

Semantic Metrics

Embedding-based similarity for deep meaning alignment.

BERTScore token-level F1
Cosine sentence similarity
Local WASM embeddings (Xenova)

LLM Judge

Mistral grades qualitative dimensions no metric can see.

Intent alignment
Completeness & grounding
Hallucination detection
Professionalism & safety

Deterministic Rule Engine

8 hard rules that every generated reply must pass — no exceptions. Violations trigger an automatic critique-and-rewrite loop.

✓ Correct dates only
✓ No fabricated names
✓ Numbers verified
✓ Attachments consistent
✓ All questions addressed
✓ Action items present
✓ No redundant questions
✓ No verbatim repetition
SELF-CORRECTION
Fails automatically.Fixes itself.

When a draft scores below 0.80, MailSense generates a structured critique, rewrites the reply, and re-evaluates — up to 2 times — before giving up or passing.

1.
Score below threshold (0.80)
Evaluator flags failed metrics and rule violations
2.
Critique generated
LLM writes structured feedback on exactly what failed
3.
Response rewritten
Draft is regenerated incorporating the critique
4.
Re-evaluated
Full evaluation runs again. Pass → saved to Gmail Drafts
Draft generated
Evaluating — 11 metrics
Score: 0.71 — FAILED0.71
Generating critique…
Rewriting draft (attempt 1)
Re-evaluating…
Score: 0.94 — PASSED ✓0.94
Saved to Gmail Drafts
FEATURES
Everything in oneworkspace.
01
OAuthGmail APIComposio

Gmail Integration

OAuth & Thread Fetching

Securely connects to your Gmail via Composio OAuth, automatically pulls unread threads, and saves evaluated drafts back to your drafts folder.

Learn more
02
Vector DBRAGEmbeddings

Pinecone Vector Search

Few-Shot Context Retrieval

Finds the top-3 most similar email pairs from your dataset using semantic search to provide relevant few-shot examples for generation.

Learn more
03
LLMMistralPrompting

Mistral AI Generation

Context-Aware Drafting

Generates professional email replies using your active prompt template, retrieved examples, and tone policy for consistent quality.

Learn more
04
BLEUROUGERules

Evaluation Engine

11-Metric Scoring Suite

Scores every draft across BLEU, ROUGE, BERTScore, LLM Judge, and 8 deterministic rules. Fails trigger automatic self-correction.

Learn more
LIVE EVALUATIONS
Scoring emailsin real time.

Every email response that flows through MailSense is evaluated, critiqued, and refined — completely autonomously.

3,837emails evaluated
Live FeedActive
gmail-fetch
Fetching unread email threads from Gmail
running1m 46s
vector-search
Searching Pinecone for similar email pairs
running3m 52s
mistral-generate
Generating reply draft with Mistral AI
running5m 48s
evaluation-score
Scoring draft across 11 quality metrics
done7m 39s
rule-check
Running deterministic rule validation
done9m 19s
Intent Alignment
BERTScore
ROUGE-L
BLEU-4
METEOR
Hallucination Detection
Grounding
Professionalism
Safety
Rule Engine
Intent Alignment
BERTScore
ROUGE-L
BLEU-4
METEOR
Hallucination Detection
Grounding
Professionalism
Safety
Rule Engine
Intent Alignment
BERTScore
ROUGE-L
BLEU-4
METEOR
Hallucination Detection
Grounding
Professionalism
Safety
Rule Engine
Gmail OAuth
Composio
Pinecone Vector Search
Mistral AI
Supabase Postgres
ONNX Transformers.js
Few-Shot RAG
Self-Correction Loop
Prompt Versioning
Audit Logs
Gmail OAuth
Composio
Pinecone Vector Search
Mistral AI
Supabase Postgres
ONNX Transformers.js
Few-Shot RAG
Self-Correction Loop
Prompt Versioning
Audit Logs
Gmail OAuth
Composio
Pinecone Vector Search
Mistral AI
Supabase Postgres
ONNX Transformers.js
Few-Shot RAG
Self-Correction Loop
Prompt Versioning
Audit Logs

Ship better email replies,
starting today.

Connect your Gmail, upload your dataset, and let MailSense generate, score, and save production-grade replies automatically.