Companies participating in complex public auctions – Auction Team
01. Experience (December 01, 2024 - Present)
Form Filling Expert – Dendrio
- Automatically identifies the required forms in the award documentation and determines the order of prioritization of completion.
- Builds and completes compliance matrices, significantly reducing the manual effort of the bidding team.
- Extract relevant data from previous offers, internal databases and public sources to auto-populate documents.
- Generate narrative sections — technical memos, methodologies — based on the organization's templates and project history.
02. Education and Training (in progress)
Advanced training in auction teams from multiple fields
- Experience gained from various projects in IT and construction, with active interdepartmental collaboration.
- Continuous update with the specifics of each tender team and the documentation they collaborate with.
03. Personal Skills
Multilingual
- Romanian, English, French, Spanish, Chinese, Arabic — documentation without language barriers.
Multi-format management
- Processes documents, spreadsheets, PDFs and scanned images, including poor quality documents through its own OCR.
Speed and volume
- Manages over 500 submissions with strict deadlines, without processing errors.
Conformity
- Integrated legislative knowledge — automatically checks documentation for alignment with applicable legal requirements.
04. Technical Architecture
Nenos Bidding AI Agent Documentation Expert (Form Completion) is built around a document processing pipeline that transforms the award documentation into structured, actionable data. For native digital documents (PDF, Word, Excel), extraction is performed directly through structural parsing. For scanned or poor quality documents — a frequent situation in auctions with digitized physical documentation — the agent uses an in-house developed OCR system, with an accuracy rate exceeding 95%, specifically optimized for formats and fonts found in Romanian and European public procurement documentation. The extracted data is normalized and introduced into the LangGraph agent flow that identifies the necessary forms, builds compliance matrices and triggers automatic completion by correlation with the database of previous bids, indexed by LlamaIndex.
A dedicated narrative section generation module — technical memos, methodology descriptions, quality criteria responses — produces structured and coherent content based on the organization’s templates and data from previous projects, significantly reducing the bidding team’s writing time. The microservices architecture independently manages the OCR service, the parsing service, the completion engine, and the audit trail service, each scaling according to the volume of active bids. Integration with SharePoint, Amazon S3, and SEAP/SICAP platforms is achieved through MCP and REST API, and on-premises deployment is available for organizations with strict confidentiality requirements regarding bid documentation.
Agent security is treated as a stand-alone architectural layer. Uploaded bidding documents represent an attack vector specific to this agent. The document sanitization service analyzes each file before processing, and uploaded document injection tests are a mandatory part of the validation cycle for each version. Each instance undergoes a rigorous adversarial testing process before delivery, covering the following vectors: fictional framing with meta-instruction, few-shot poisoning, roleplay with identity transfer, uploaded document injection, chained context manipulation, linguistic obfuscation through similar characters or invisible spaces, tool/API injection, as well as full coverage of the OWASP LLM Top 10 vulnerabilities (prompt injection, insecure output handling, training data poisoning, model denial of service, supply chain vulnerabilities and others).