Landry Bodjona

Landry Bodjona · Full-stack and AI engineering

Full-stack engineer · AI systems · Product engineering

I build AI systems you can operate, inspect, and trust.

I build full-stack applications, AI agents and reliable workflows. Explore ten working demos, inspect their traces and see the engineering decisions behind them.

Explore selected work Read this trace
PythonDjangoFlaskNode.jsTypeScriptNuxt / Vue

All ten systems builtNo accounts · English and Czech · Free model tiers behind one gateway

Selected engineering workThe Scope

Ten systems, one platform. Each tile is a machine at work.

Explore all ten systems

10 / systems

LB-01 Support Desk Agent · a recorded run

The Scope

Explore the system
The ScopeLB-01 Support Desk Agent · a recorded runReplay

9 steps and 5 model calls in 10.9 seconds, recorded on 2026-10-07.

  1. 01 redact PIIunder 10 ms
  2. 02 screen for injectionllama-prompt-guard-2-86m · 43 in, 0 out
  3. 03 classifygpt-oss-20b · 442 in, 169 out
  4. 04 hybrid searchbge-m3
  5. 05 rerankbge-reranker-base
  6. 06 look up orderunder 10 ms
  7. 07 draft with citationsgpt-oss-120b · 866 in, 957 out
  8. 08 check claimsunder 10 ms
  9. 09 routeunder 10 ms
ModelToolRetrievalGuard5 model calls · 9 steps · 10.9 s · recorded 2026-10-07
10working AI systems
3backend stacks
EN / CZlanguage options
↗inspectable execution traces

Selection guide

Filter the ten systems by back end or technique, the way engineers search a parts distributor. Select a part number to open its datasheet.

Showing 10 of 10 systems
The ten systems with their back end, techniques and build phase
PartSystemWhat a visitor doesBack endTechniquesPhase
LB-01 Support Desk Agent Files a ticket, then approves the cited replyDjango · RAG · Tool use · CitationsFiles a ticket, then approves the cited replyDjangoRAG · Tool use · Citations1
LB-02 Booking Concierge Books a tasting in chat, then tries to double-book itDjango Channels · Tool use · Real-timeBooks a tasting in chat, then tries to double-book itDjango ChannelsTool use · Real-time2
LB-03 Invoice Reader Uploads an invoice, fixes a field, exports entriesFlask · async · Vision · Structured outputUploads an invoice, fixes a field, exports entriesFlask · asyncVision · Structured output2
LB-04 Contract Radar Reviews a contract and jumps to each risky clauseNode + TS · Citations · Structured outputReviews a contract and jumps to each risky clauseNode + TSCitations · Structured output2
LB-05 Data Analyst Asks a sales question, then tries to drop a tableFlask · sync · Tool use · Structured outputAsks a sales question, then tries to drop a tableFlask · syncTool use · Structured output1
LB-06 Incident Commander Breaks the shop and approves the agents’ fixNode + TS · Multi-agent · Tool use · Real-timeBreaks the shop and approves the agents’ fixNode + TSMulti-agent · Tool use · Real-time3
LB-07 QA Engineer Switches bugs on, gets a bug report and a testNode + TS · Browser agent · Tool useSwitches bugs on, gets a bug report and a testNode + TSBrowser agent · Tool use2
LB-08 Automation Studio Describes a process, runs it, then makes it failNode + TS · Structured output · Tool useDescribes a process, runs it, then makes it failNode + TSStructured output · Tool use1
LB-09 Meeting Recorder Records a minute, clicks an action to hear itDjango · Voice · Local model · Structured outputRecords a minute, clicks an action to hear itDjangoVoice · Local model · Structured output3
LB-10 Eval Lab Edits a prompt, compares score and costFlask · async · Evals · Multi-providerEdits a prompt, compares score and costFlask · asyncEvals · Multi-provider3

What every system page shows

The datasheet
What the system does, the problem it solves and its operating limits, in a 30-second Brief or the full Technical version.
The evaluation board
The live demo. It opens on curated samples with cached results, and custom input runs for real within the visitor’s daily quota.
The Scope
A trace of every run: each step, tool call, token and model, with its latency and the provider that answered.

Build order

  1. Phase 1

    Foundation

    • LB-00Platform
    • LB-01Support Desk Agent
    • LB-05Data Analyst
    • LB-08Automation Studio

    These touch all three back ends, the gateway, tracing, quotas and the eval harness. Everything after reuses them.

  2. Phase 2

    Breadth

    Real-time chat, vision, long documents and browser automation, on a stable base.

  3. Phase 3

    Showpieces

    The most complex builds come last. Eval Lab arrives with golden sets that have been growing since Phase 1.

About · Engineering mindset

Built with curiosity. Shipped with discipline.

I'm Landry Bodjona, a full-stack and AI engineer. I design usable products and take care of the details behind the interface: architecture, backend workflows, testing, observability and the experience of the person using it. These ten systems are interactive demonstrations using a reference business and demo data.

Review my GitHub work

Let's build something useful

Need an engineer who can build the whole thing?

Open to remote engineering projects and technical collaborations. Explore the code, test the systems and start a conversation.

Connect on GitHub