Conversational analysis
A streaming AI backend supporting Claude, OpenAI and Gemini, with provider fallback and persistent multi-file sessions.
Analytics DepotFrom business data to a useful conversation.
A B2B analytics platform that lets users ask questions about business data and uploaded files. Waleed Abbas built the AI experience and the production systems around it as Analytics Depot’s Founding AI Engineer.
- Delivery role
- Waleed Abbas · Founding AI Engineer
- Project context
- B2B analytics SaaS
- Focus
- AI agents, document intelligence and data integrations
One interface for questions, files and connected data.
The product needed to connect a conversational interface with business datasets, uploaded documents and analysis tools. The engineering work covered both the AI interaction and the operational requirements of a production SaaS platform.
What the delivered system supports
Users can bring a question and its data context into one conversational workflow. The delivered scope connects file retrieval, analysis tools and database access with the application infrastructure needed to operate the product.
Connected by design- Business sources
Uploaded files and connected datasets
- Retrieval and analysis
Relevant context, tool execution and validated queries
- A conversational answer
Streaming analysis with persistent session context
What was delivered.
Document intelligence
File processing, retrieval-based search and summaries over uploaded documents, bringing relevant source context into the conversation.
Natural-language-to-SQL
A path from a business question to a database query, with schema retrieval, validation and safety checks around execution.
Connected production infrastructure
Data connectors, background workers, subscription billing, database migrations, automated tests and production deployment.
From requirement to implemented capability.
Selected engineering scope, with the boundaries that matter when evaluating a similar project.
| Requirement | Implemented capability | Boundary |
|---|---|---|
| Ask questions across files and datasets | Persistent multi-file sessions, document retrieval and data connectors. | Retrieval provides context; outputs still need to be assessed against the source. |
| Turn a question into database analysis | Schema retrieval, validated natural-language-to-SQL and sandboxed analysis. | Validation addresses execution risks; it does not establish that every answer is correct. |
| Operate the AI features inside a SaaS product | Streaming multi-provider backend, Celery/Redis workers, migrations, tests and deployment. | This case documents delivered scope, not a measured revenue or accuracy improvement. |
“He scopes problems, proposes solutions, ships them, and monitors them in production.”
Explore the work.
This case study describes Waleed Abbas’s work in his Founding AI Engineer role at Analytics Depot. The written testimonial is attributed to Analytics Depot’s CEO.
- Read the client’s testimonial PDF · opens in a new tab
- Watch a product query demo Google Drive · opens in a new tab
- Watch the web-search demo Google Drive · opens in a new tab
Bring one real workflow.
We’ll map the useful starting point, the tools involved and the checks it needs.