Social Media Data — Extraction and analysis
Structured social data serving your commercial intelligence.
Goal
Get weekly structured social data to drive your marketing decisions.
How ·Multi-network ETL collection respecting ToS/GDPR + normalization + data warehouse delivery + recurring reporting.
Structured collection of public data from social networks in strict compliance with terms of use and GDPR. Cleaned, normalized data exported to your analytics tools or data warehouse. MCP-first architecture orchestrated through our ATLAS-Agentic methodology.
Key features
- Structured extraction from social APIs and public sources
- Automatic normalization and cleaning
- Sentiment and trend analysis
- Export to data warehouse or analytics tools
- Compliance with GDPR and platform terms of use
Technologies
Social APIs, ETL orchestration, data warehouse.
Desktop AI Terminal — Unified local interface
Analyst productivity on local workstation, sensitive contexts mastered.
Goal
Give your analysts the power of AI without your sensitive data transiting through the cloud.
How ·Desktop app integrating cloud and local LLMs + business tools + confidentiality preserved on sensitive contexts.
Desktop application integrating multiple AI models and business tools in a unified interface. Local confidentiality for contexts where data must not transit through the cloud.
Key features
- Multi-model desktop interface (cloud and local)
- Local confidentiality for sensitive data
- Business tool integration (CRM, ERP, document base)
- History and resumption of work sessions
- Per-user profile customization
Technologies
Desktop application, cloud and local AI models, API integrations.
13
In progress / availableEnterprise document RAG
AI-augmented search on your document heritage — answers grounded in your documents, with sourced citations.
Goal
Find precise information in your document estate by asking a question, with citation of the source document.
How ·Multi-source vector indexing + LLM with guardrails + journaling + fine-grained access control.
RAG (Retrieval-Augmented Generation) platform connected to your internal sources (legal, HR, technical, contracts, regulatory, finance, accounting). Sourced responses with document citations, complete logging, access control, quality and confidentiality audits. Cross-functionally deployable across all enterprise functions.
Key features
- Vector indexing of your document heritage
- Sourced responses with verifiable citations
- Role-based access control and per-document confidentiality
- Complete logging of prompts and responses
- Continuous quality bench and drift detection
Technologies
LLM (Claude, GPT, Mistral), pgvector, Pinecone, custom indexing, orchestration with guardrails.
Lotus Notes / HCL Domino estate recovery
Recover and make usable a Lotus mail history that nobody in the company can open any more.
Goal
Recover the decision history locked inside a Lotus mail system before switching it off.
How ·Extraction through the Domino server + intact attachment recovery + automatic classification by business topic.
Many companies keep a Lotus Notes / HCL Domino server on life support for one reason only: years of exchanges, decisions and attachments are locked inside it. The usual reflex — grab the `.nsf` files and open them elsewhere — is a dead end: **the content of an `.nsf` is encrypted by Domino and holds no usable text outside the server**. Our approach starts from the opposite fact: the server itself renders the decrypted content to an authenticated user. We therefore extract the estate through that path — messages, views and attachments — then classify it by business topic to turn it back into usable material: a searchable archive, a knowledge base, or an input corpus for document RAG.
Key features
- Extraction through the Domino server (authenticated session), with no dependency on the encrypted .nsf format
- Reading of database views and automated traversal of received and sent messages
- Recovery of each message's readable body, stripped of webmail markup
- Download of intact, openable attachments (PDF, office documents, archives)
- Automatic classification of messages by business topic, flagging exchanges with commercial or contractual stakes
Technologies
HCL Domino / iNotes, HTTP extraction over an authenticated session, traversal of views and documents, attachment recovery, automatic topic classification.
26
In progress / availableEl-Facture — Invoicing, purchases and bank statements for small businesses
Issue your invoices, read your suppliers' invoices and match your bank statement in one place, then hand everything to your accountant.
Goal
Give a small business and its accountant clean, current documents without installing an ERP.
How ·Online application + OCR on purchase invoices + PDF bank statement import + matching + withholding tax + FEC export + invoice creation from a WhatsApp voice or text message, with a draft to confirm.
El-Facture is an online application published by our group that gathers what a small business issues and receives every month: the sales invoices it sends, the purchase invoices it receives (read by OCR), the bank statements it imports as PDF and matches, the withholding tax prepared for filing, and the standard ledger export (FEC) it hands to its accountant. Separate from these functions in service, one capability is connected at scoping: El-Facture is designed to sign invoices electronically according to the rules of each country: the accredited signature provider of the country is connected at scoping. El-Facture is a facilitator between the company and its accountant, not an ERP and not bookkeeping software. The application is in production; no client uses it in day-to-day production yet, and a first pilot with accounting firms is scheduled. Product site: [el-facture.com](https://el-facture.com).
Key features
- Sales invoices and credit notes, continuous numbering, due-date tracking and overdue alerts
- Purchase invoices dropped as PDF or photo and read by OCR: supplier, dates, amounts and taxes extracted, then confirmed by the user
- PDF bank statement import (three banks validated on our internal test bench) and matching against invoices issued and received
- Withholding tax computed and exported in the XML format expected by the local tax administration, plus the standard accounting ledger export (FEC, required in France) for the accountant
- Invoice creation from WhatsApp, by voice or text message: a PDF draft is sent back for confirmation before the invoice is issued
Technologies
Hosted web application (app.el-facture.com), OCR on purchase invoices, PDF bank statement parsing, XML and FEC exports, invoice creation from a WhatsApp message (voice or text) with a PDF draft to confirm. Electronic signature through the accredited provider of each country, connected at scoping.
On-premise contract analysis on a local server — Clause extraction with no data leaving your premises
Read a batch of contracts and pull out the clauses that matter, with an AI model running on your own hardware.
Goal
Get a reliable table of the key clauses across a batch of contracts without a single page leaving the company network.
How ·Open language model installed locally + fixed 10-field grid + a strict rule to answer "not stated" rather than guess + human review.
A 30-page contract holds about ten facts that a management team actually needs: who signs, for what, from when, for how much, on what payment terms, with what notice period, under which law. This tool extracts them automatically, contract by contract, and lays them out in a table. What sets it apart: the AI model is **installed on a machine inside the company** (an open model run with Ollama). The text the model analyses is never sent to an online service. It was built for legal and finance departments that, rightly, refuse to hand their contracts to an external AI provider. The tool is currently at the **internal test** stage: it has been evaluated on a batch of 11 contracts and is not yet installed at any client. Deployment starts with a trial on your own contracts.
Key features
- Open language model executed on a local machine (Ollama, Gemma 4): contracts stay inside the company network
- Fixed grid of 10 fields per contract: parties, subject, signature date, effective date and term, amounts, payment terms, termination and notice, governing law and jurisdiction, confidentiality, penalties
- Strict rule: if the information is not in the contract, the field reads "not stated", never a guessed value
- Handles digital contracts in French and English (the test batch included one English contract); local OCR for scanned contracts is set at scoping
- Output as structured JSON per contract, consolidated into a working table (one row per contract, one column per field) ready for review and sorting
Technologies
Ollama running an open model (Gemma 4) on a local machine; 10-field extraction grid; JSON output per contract, consolidated into a table; local OCR for scans set at scoping. No graphical interface at this stage (command-line execution).
29
In progress / availableExport file checker — Cross-checking shipping documents before departure
Re-read a complete export file, document by document, and flag every inconsistency before the container leaves.
Goal
Catch, before shipment, the discrepancies between the purchase order, invoices, packing lists, certificates and bill of lading of an export file.
How ·OCR and vision on native or scanned PDFs + schema-based extraction citing its source + cross-check rules (the purchase order is the reference) + HTML report + correction e-mails drafted, never sent.
An export file brings together about ten documents issued by different parties: the customer's purchase order, invoices, packing lists, certificates of origin, analysis, health or phytosanitary certificates, the bill of lading. One quantity that differs between invoice and packing list, one mistyped container number, one missing certificate, and the goods are held at the port or payment is delayed. This tool reads every document (digital or scanned), extracts the key data while citing exactly where it read it, then runs a set of cross-checks built around one simple rule: **the purchase order is the reference**. The output is an HTML report with a built-in document viewer to verify each alert, plus correction request e-mails drafted per issuer, in French and English, that the exporter reviews and sends personally. The project is at **pilot** stage: it was delivered as source code to an exporter, and day-to-day use by the client is not yet established.
Key features
- Reads every document in a file: purchase order, invoices, packing lists, certificates of origin, analysis, health and phytosanitary certificates, bills of lading, as native or scanned PDF (OCR and vision)
- Schema-based extraction with two reconciled reading paths and source citation: every extracted value points back to the document and the spot where it was read
- Cross-checks: the purchase order is the reference, required-document list per file, configurable tolerances, partial-shipment handling, internal consistency of each document
- Container number check digit verification (ISO 6346) to catch typing errors
- HTML report with built-in viewer: each alert opens on the relevant page of the document
Technologies
Mistral OCR and language model, hosted in the European Union (extraction on the text of every document); schema-based extraction reconciling two reading paths; cross-check rules engine; standalone HTML report with built-in viewer; 408 automated tests.