Develop Automated Tender Discovery and CRM System
Budget: €30 – €250 EUR
PHASE 1 – Automated Tender Discovery + CRM + AI Snapshot
Goal: Fully automated ingestion + structured CRM + AI tender analysis.
Preferred / Required Tech Stack (Phase 1):
Frontend: React + TypeScript (Next.js preferred)
Backend/API: Node.js (NestJS/Express) or Python (FastAPI)
Database: PostgreSQL (mandatory)
Vector store: pgvector (inside PostgreSQL)
AI: OpenAI API (must be modular so we can swap provider later)
File storage: S3-compatible object storage (AWS S3 / Cloudflare R2 / similar)
OCR: Tesseract (open-source) acceptable for MVP
Scheduler: cron/worker-based job runner (runs 09:00 / 12:00 / 15:00 GMT)
Deployment: Dockerised (mandatory)
1. Saved Search Presets
Users define:
Category / CPV codes
Keywords
Counties (multi-select, all 26)
Tender type filters
Presets must be savable and schedulable.
2. Scheduled Tender Pull Engine
System runs at:
09:00 GMT
12:00 GMT
15:00 GMT
For each preset:
Pull new/updated tenders
Create/update CRM record
Prevent duplicates
3. Tender CRM Record Fields
Title
Authority
Reference ID
Location
Estimated value
Submission deadline
Project start date
Status:
New
Reviewing
Bidding
Submitted
Won
Lost
Dismissed
4. Document Ingestion Engine
Auto-download publicly available documents.
Attempt logged-in retrieval for restricted docs.
If retrieval fails → flag as “Login Required – Pending”.
Store all documents in structured tender vault.
5. AI Tender Snapshot
Auto-generate:
3-line summary
Key criteria checklist
Timeline extraction
Risk flags
Recommended actions
Structured internal output required.
6. Alerts System
14 / 7 / 3 / 1 day deadline alerts
Dashboard warnings
“No activity” reminders
7. AI Query Console (Phase 1 Basic)
Firm-level queries such as:
“How many tenders due next 7 days?”
“Show refurbishment tenders in Dublin.”
“Highest value opportunity this month.”
Must use RAG over CRM + stored documents.
Deliverable:
Fully automated discovery + import + analysis + alerts.
PHASE 2 – Bid Workspace + Compliance Engine
Bid workspace per tender
Task assignment
Compliance checklist extraction
Missing requirements detection
Proposal drafting engine
Schedule of rates library
Long-form document generation matching tender structure
PHASE 3 – Win/Loss Intelligence Engine
Record outcome (Won/Lost)
Feedback storage
AI pattern detection
Win-rate analytics by category/authority
Reusable winning answer library
PHASE 4 – Advanced Intelligence Layer
Authority profiling
Framework tracking
Opportunity scoring
Risk scoring model
Suggested best-fit tenders
PHASE 5 – Enterprise & Deployment
Dedicated deployment option
Role matrix expansion
MFA
Audit logs
Encryption at rest
Export/import tools
PHASE 6 – Integrations & Advanced Automation
Email integration
Calendar integration
Multi-company bidding support
Advanced drafting refinement
External collaboration options
7. AI Architecture Expectations
Must include:
Proper RAG pipeline
Controlled chunking
Embedding once per document
Top-k retrieval limits
Structured output
Token limits
Caching layer
AI must NOT:
Reprocess full tender packs repeatedly
Generate uncontrolled verbose outputs
Ignore cost control
8. Developer Instructions
Quote EACH phase separately.
Phase 1 will be awarded first quote for all phases now just phase 1.
Subsequent phases awarded based on quality.
Developer must clearly explain ingestion strategy.
Clean documentation required.
Goal: Fully automated ingestion + structured CRM + AI tender analysis.
Preferred / Required Tech Stack (Phase 1):
Frontend: React + TypeScript (Next.js preferred)
Backend/API: Node.js (NestJS/Express) or Python (FastAPI)
Database: PostgreSQL (mandatory)
Vector store: pgvector (inside PostgreSQL)
AI: OpenAI API (must be modular so we can swap provider later)
File storage: S3-compatible object storage (AWS S3 / Cloudflare R2 / similar)
OCR: Tesseract (open-source) acceptable for MVP
Scheduler: cron/worker-based job runner (runs 09:00 / 12:00 / 15:00 GMT)
Deployment: Dockerised (mandatory)
1. Saved Search Presets
Users define:
Category / CPV codes
Keywords
Counties (multi-select, all 26)
Tender type filters
Presets must be savable and schedulable.
2. Scheduled Tender Pull Engine
System runs at:
09:00 GMT
12:00 GMT
15:00 GMT
For each preset:
Pull new/updated tenders
Create/update CRM record
Prevent duplicates
3. Tender CRM Record Fields
Title
Authority
Reference ID
Location
Estimated value
Submission deadline
Project start date
Status:
New
Reviewing
Bidding
Submitted
Won
Lost
Dismissed
4. Document Ingestion Engine
Auto-download publicly available documents.
Attempt logged-in retrieval for restricted docs.
If retrieval fails → flag as “Login Required – Pending”.
Store all documents in structured tender vault.
5. AI Tender Snapshot
Auto-generate:
3-line summary
Key criteria checklist
Timeline extraction
Risk flags
Recommended actions
Structured internal output required.
6. Alerts System
14 / 7 / 3 / 1 day deadline alerts
Dashboard warnings
“No activity” reminders
7. AI Query Console (Phase 1 Basic)
Firm-level queries such as:
“How many tenders due next 7 days?”
“Show refurbishment tenders in Dublin.”
“Highest value opportunity this month.”
Must use RAG over CRM + stored documents.
Deliverable:
Fully automated discovery + import + analysis + alerts.
PHASE 2 – Bid Workspace + Compliance Engine
Bid workspace per tender
Task assignment
Compliance checklist extraction
Missing requirements detection
Proposal drafting engine
Schedule of rates library
Long-form document generation matching tender structure
PHASE 3 – Win/Loss Intelligence Engine
Record outcome (Won/Lost)
Feedback storage
AI pattern detection
Win-rate analytics by category/authority
Reusable winning answer library
PHASE 4 – Advanced Intelligence Layer
Authority profiling
Framework tracking
Opportunity scoring
Risk scoring model
Suggested best-fit tenders
PHASE 5 – Enterprise & Deployment
Dedicated deployment option
Role matrix expansion
MFA
Audit logs
Encryption at rest
Export/import tools
PHASE 6 – Integrations & Advanced Automation
Email integration
Calendar integration
Multi-company bidding support
Advanced drafting refinement
External collaboration options
7. AI Architecture Expectations
Must include:
Proper RAG pipeline
Controlled chunking
Embedding once per document
Top-k retrieval limits
Structured output
Token limits
Caching layer
AI must NOT:
Reprocess full tender packs repeatedly
Generate uncontrolled verbose outputs
Ignore cost control
8. Developer Instructions
Quote EACH phase separately.
Phase 1 will be awarded first quote for all phases now just phase 1.
Subsequent phases awarded based on quality.
Developer must clearly explain ingestion strategy.
Clean documentation required.