Sales Engineering Agentic System

An AI-powered multi-agent pipeline that automates the sales engineering workflow: ingest RFP/RFQ documents, extract structured intelligence using Google Gemini, match the best-fit employees from your talent pool, and send outreach notifications — all orchestrated through a modern web interface.

Architecture

flowchart TB
    subgraph Frontend ["Next.js Frontend (:3000)"]
        UI[Web UI - React 19]
        API_ROUTES[Next.js API Routes]
        STORAGE[Local JSON Storage]
    end

    subgraph Backend ["FastAPI Backend (:8000)"]
        ANALYZE["/api/analyze"]
        ANALYZE_MULTI["/api/analyze-multi"]
        MATCH["/api/match"]
        EXTRACT_CV["/api/extract-cv"]
    end

    subgraph AI ["Google Gemini 2.5 Flash"]
        RFP_AGENT[RFP Analysis Agent]
        MATCHMAKER[Matchmaking Agent]
        CV_PARSER[CV Extraction]
    end

    subgraph External ["External Services"]
        NTFY[ntfy.sh Notifications]
        GDRIVE[Google Drive]
        GSHEETS[Google Sheets CRM]
    end

    UI -->|Upload RFP/RFQ| API_ROUTES
    API_ROUTES -->|Forward files| ANALYZE_MULTI
    API_ROUTES -->|Employee matching| MATCH
    API_ROUTES -->|CV upload| EXTRACT_CV
    API_ROUTES -->|Send notifications| NTFY
    API_ROUTES <-->|Read/Write| STORAGE

    ANALYZE --> RFP_AGENT
    ANALYZE_MULTI --> RFP_AGENT
    MATCH --> MATCHMAKER
    EXTRACT_CV --> CV_PARSER

    GDRIVE -->|CV ingestion| CV_PARSER
    CV_PARSER -->|Structured data| GSHEETS

Pipeline Flow

stateDiagram-v2
    [*] --> INGESTION: Upload RFP/RFQ documents
    INGESTION --> ANALYSIS: Gemini extracts requirements, skills, risks
    ANALYSIS --> PLANNING: Human clarification + employee matching
    PLANNING --> OUTREACH: PM selected, team assembled
    OUTREACH --> [*]: Notifications sent via ntfy.sh

Ingestion — User uploads one or more RFP/RFQ documents (PDF, DOCX, TXT)
Analysis — Gemini 2.5 Flash extracts structured intelligence: requirements, skills matrix, deadlines, risks, compliance norms
Planning — System matches employees against extracted requirements, human provides clarification if needed, PM is auto-selected
Outreach — Notifications are sent to the PM and matched team members via ntfy.sh

Tech Stack

Layer	Technology	Purpose
Frontend	Next.js 16, React 19, TypeScript	App Router-based web interface
Styling	Tailwind CSS v4, Framer Motion	Responsive design and animations
Backend API	FastAPI, Uvicorn	Python AI/ML service layer
Backend Orchestration	Next.js API Routes	Request orchestration, state management
AI/ML	Google Gemini 2.5 Flash	Structured extraction, matchmaking, CV parsing
Document Parsing	pdfplumber, python-docx, Gemini Files API	PDF/DOCX/TXT text extraction (with multimodal fallback)
Data Storage	Local JSON files	Employee and project persistence
Notifications	ntfy.sh	Push notifications to team members
Onboarding	Google Drive API, Google Sheets API	CV ingestion pipeline to CRM
Validation	Pydantic v2	Structured output validation and schemas

Prerequisites & Setup

Requirements

Python 3.10+
Node.js 18+
Gemini API Key — get one here

1. Clone & configure

git clone <repo-url>
cd web

# Create your .env file
cp .env.example .env
# Set your Gemini API key:
#   GEMINI_API_KEY=your_key_here

2. Install dependencies

# Python dependencies (FastAPI + RFP agent + AI libraries)
pip3 install -r requirements.txt

# Node.js dependencies (Next.js frontend)
npm install

3. Run — open two terminals

Terminal 1 — Python API server (port 8000):

python3 -m uvicorn server:app --port 8000

Terminal 2 — Next.js frontend (port 3000):

npm run dev

Open http://localhost:3000 in your browser.

Note: Both servers must be running simultaneously. The frontend proxies AI requests to the Python backend.

macOS Troubleshooting

Use python3 / pip3 — macOS ships with Python 2 by default
If uvicorn not found: python3 -m uvicorn server:app --port 8000
Port conflict: try --port 8001 and set NEXT_PUBLIC_AGENT_API_URL=http://localhost:8001 in .env

Environment Variables

Variable	Required	Default	Description
`GEMINI_API_KEY`	Yes	—	Google Gemini API key for all AI operations
`NEXT_PUBLIC_AGENT_API_URL`	No	`http://localhost:8000`	URL of the Python FastAPI server
`CRM_SPREADSHEET_ID`	For onboarding	—	Google Sheets spreadsheet ID for CRM
`DROPZONE_FOLDER_NAME`	For onboarding	`01_CV_Dropzone`	Google Drive folder name for incoming CVs
`PROCESSED_FOLDER_NAME`	For onboarding	`02_Processed_CVs`	Google Drive folder for processed CVs
`SKILLS_DB_SHEET_NAME`	For onboarding	`Skills_DB`	Sheet tab name in the CRM spreadsheet
`GEMINI_MODEL`	No	`gemini-1.5-flash`	Model override for onboarding agent

Running the Application

Web Interface (recommended)

Start both servers (see Setup above)
Navigate to http://localhost:3000
Upload an RFP/RFQ document
Review extracted requirements, skills, and risks
Provide clarification when prompted
View matched employees and selected PM
Trigger outreach notifications

Demo Walkthrough

Upload — Drag and drop any RFP PDF into the upload zone
Analysis — The system displays extracted requirements, skills matrix, deadlines, and risks within seconds
Matchmaking — Answer the clarification question, then view ranked employee matches with fit scores
Outreach — Click to notify the selected PM and team via push notifications

API Documentation

Python FastAPI Endpoints (port 8000)

`POST /api/analyze`

Analyze a single uploaded RFP document.

Field	Value
Content-Type	`multipart/form-data`
Body	`file` — PDF, DOCX, or TXT

Response:

{
  "requirements": ["REQ-001 [HIGH*] SCADA integration: ..."],
  "ambiguity": "Scope boundary between Phase 1 and Phase 2 is unclear...",
  "suggestedSkills": ["Siemens TIA Portal — senior × 2"],
  "rfp_title": "...",
  "client_name": "...",
  "rfp_summary": "...",
  "submission_deadline": "2025-06-30",
  "project_duration": "18 months",
  "confidence_score": 0.87,
  "deadlines": [...],
  "risks": [...],
  "dependencies": [...],
  "compliance_norms": [...],
  "key_evaluation_criteria": [...],
  "pitfalls": [...],
  "planning_payload": { ... }
}

`POST /api/analyze-multi`

Analyze multiple uploaded files as one combined RFP.

Field	Value
Content-Type	`multipart/form-data`
Body	`files` — multiple PDF/DOCX/TXT files

Response: Same schema as /api/analyze.

`POST /api/extract-cv`

Extract structured employee profile from a CV PDF.

Field	Value
Content-Type	`multipart/form-data`
Body	`file` — PDF CV

Response:

{
  "status": "success",
  "data": {
    "firstName": "Anna",
    "lastName": "Schmidt",
    "email": "anna@example.com",
    "location": "Munich",
    "role": "Automation Engineer",
    "level": "senior",
    "yearsOfExperience": 8,
    "pastIndustryExperience": ["Energy", "Manufacturing"],
    "skills": ["Siemens TIA Portal", "SCADA", "IEC 61850"],
    "certifications": ["ISO 9001 Auditor"],
    "linkedin": "https://linkedin.com/in/anna-schmidt"
  }
}

`POST /api/match`

Match employees against project requirements using Gemini.

Field	Value
Content-Type	`application/json`

Request:

{
  "requirements": ["REQ-001 [HIGH*] SCADA integration: ..."],
  "employees": [
    {
      "id": "emp-1",
      "name": "Anna Schmidt",
      "role": "Automation Engineer",
      "skills": ["Siemens TIA Portal", "SCADA", "IEC 61850"]
    }
  ],
  "human_answer": "optional clarification string"
}

Response:

{
  "matches": [
    {
      "id": "emp-1",
      "name": "Anna Schmidt",
      "role": "Automation Engineer",
      "match": 92,
      "reason": "Direct match on all three core technical skills.",
      "skills": ["Siemens TIA Portal", "SCADA", "IEC 61850"]
    }
  ]
}

`GET /health`

Health check endpoint. Returns { "status": "ok" }.

Next.js API Routes (port 3000)

`POST /api/process-rfq`

Orchestrates RFP ingestion: accepts files, calls Python /api/analyze-multi, creates project state.

Field	Value
Content-Type	`multipart/form-data`
Body	`files` — one or more RFP documents

Response:

{
  "status": "success",
  "project": {
    "id": "uuid",
    "stage": "ANALYSIS",
    "requirements": [...],
    "clarificationQuestion": "...",
    "thoughtLog": [...],
    "createdAt": "...",
    "updatedAt": "..."
  }
}

`POST /api/matchmaking`

Adds human clarification, triggers employee matching, auto-selects PM.

Request:

{
  "projectId": "uuid",
  "humanAnswer": "We need someone with IEC 61850 experience"
}

Response:

{
  "status": "success",
  "project": {
    "matchCandidates": [...],
    "selectedPM": { ... },
    "thoughtLog": [...],
    "status": "PLANNING"
  }
}

`POST /api/outreach`

Sends push notifications to selected PM and team members via ntfy.sh.

Request:

{
  "projectId": "uuid"
}

Response:

{
  "status": "success",
  "notifications": [...],
  "pm": { ... }
}

`GET /api/employees`

Returns the full employee list from local storage.

Response:

{
  "status": "success",
  "employees": [...]
}

`POST /api/employees/[id]`

Create or update an employee profile by ID.

Request: JSON employee object with fields matching the CandidateProfile schema.

`POST /api/employees/extract`

Upload a CV PDF, extract structured profile via Gemini, store the file, detect duplicates.

Field	Value
Content-Type	`multipart/form-data`
Body	`file` — PDF CV

Response:

{
  "status": "success",
  "employee": { ... },
  "cvUrl": "/cvs/filename.pdf",
  "matches": []
}

Core Modules

RFP Analysis Agent (`rfp_agent/`)

The core intelligence engine. Analyzes RFP/RFQ documents and extracts structured data.

How it works:

Document Loading — Supports PDF (via pdfplumber or Gemini Files API for scanned docs), DOCX (python-docx), and plain text
Chunking — For documents exceeding ~700K characters, splits into overlapping paragraph-boundary chunks
Parallel Analysis — Sends 3 parallel Gemini calls (Part A: metadata + requirements, Part B: skills + deadlines, Part C: risks + compliance) for faster processing
Merge & Deduplicate — Combines results, deduplicates requirements, renumbers IDs
Structured Output — Returns a fully validated RFPAnalysis Pydantic model

Key files:

rfp_agent/agent.py — RFPAnalysisAgent class with analyze_file() and to_planning_agent_payload()
rfp_agent/models.py — Pydantic models (RFPAnalysis, Requirement, SkillRequirement, etc.)
rfp_agent/prompts.py — System, analysis, and merge prompts for Gemini
rfp_agent/document_loader.py — Multi-format document loading with fallbacks

Matchmaking Agent

Ranks employees against extracted requirements using Gemini's reasoning capabilities.

How it works:

Receives requirements list + employee pool + optional human clarification
Sends to Gemini for semantic matching (skills, experience, certifications)
Returns ranked list with match scores (0-100) and reasoning

CV Extraction

Extracts structured CandidateProfile data from uploaded CV PDFs using Gemini.

Extracted fields: name, email, location, role, level, years of experience, industry experience, skills, certifications, LinkedIn URL.

Onboarding Agent (`onboarding_agent/`)

Automated pipeline for bulk CV ingestion from Google Drive into a CRM spreadsheet.

Flow:

Authenticates with Google APIs (Drive + Sheets)
Lists PDFs in a configured Google Drive "dropzone" folder
Downloads and parses each CV with Gemini
Appends structured data rows to a Google Sheets CRM
Moves processed files to an archive folder
Continues on per-file failures without stopping

Key files:

onboarding_agent/main.py — Pipeline orchestration
onboarding_agent/services/extractor.py — Gemini-powered CV parsing
onboarding_agent/services/google_api.py — Google Drive/Sheets client

Outreach & Notifications

Sends push notifications to team members via ntfy.sh — a simple HTTP-based pub/sub notification service. No signup required.

Data Models

RFPAnalysis (full output)

RFPAnalysis
├── rfp_title               string
├── client_name             string | null
├── rfp_summary             string          — 2-3 sentence executive summary
├── project_scope           string          — detailed scope of work
├── submission_deadline     string | null
├── project_duration        string | null
├── estimated_team_size     string | null
├── budget_constraints      string | null
├── requirements[]          Requirement
│   ├── id                  REQ-001, REQ-002, ...
│   ├── category            functional | technical | operational | compliance | financial | resource
│   ├── title               string
│   ├── description         string
│   ├── priority            high | medium | low
│   ├── is_mandatory        bool
│   ├── skills_needed       string[]
│   └── section_reference   string | null
├── skills_required[]       SkillRequirement
│   ├── skill               string (granular, e.g. "OCPP 2.0.1" not "EV charging")
│   ├── category            technical | domain | certification | soft_skill | language
│   ├── proficiency_level   junior | mid | senior | expert
│   ├── quantity_needed     int
│   ├── context             string
│   └── related_requirement_ids  string[]
├── deadlines[]             Deadline
├── dependencies[]          Dependency
├── risks[]                 Risk
├── compliance_norms[]      ComplianceNorm
├── key_evaluation_criteria string[]
├── pitfalls                string[]
├── analysis_notes          string[]
└── confidence_score        float (0.0–1.0)

Planning Payload (slim handoff)

A lightweight subset of RFPAnalysis optimized for downstream agent consumption:

{
  "rfp_title": "...",
  "client_name": "...",
  "submission_deadline": "...",
  "project_duration": "...",
  "estimated_team_size": "...",
  "skills_required": [...],
  "high_priority_requirements": [...],
  "dependencies": [...],
  "compliance_norms": [...]
}

Project States

State	Description
`INGESTION`	Documents uploaded, awaiting analysis
`ANALYSIS`	RFP analyzed, clarification question generated
`PLANNING`	Employees matched, PM selected
`OUTREACH`	Notifications sent to team

CandidateProfile / Employee

{
  "firstName": "string",
  "lastName": "string",
  "email": "string",
  "location": "string",
  "role": "string",
  "level": "junior | mid | senior | expert",
  "yearsOfExperience": "number",
  "pastIndustryExperience": ["string"],
  "skills": ["string"],
  "certifications": ["string"],
  "linkedin": "string | null"
}

CLI Usage

Run the agent directly from the terminal without the web UI:

# Single document
python3 main.py rfp.pdf

# Multiple documents
python3 main.py rfp1.pdf rfp2.docx rfp3.pdf

# Save results to files
python3 main.py rfp1.pdf rfp2.pdf --output-dir results/ --planning-dir planning/
# Produces:
#   results/rfp1.json             — full analysis
#   results/rfp2.json
#   planning/rfp1_planning.json   — slim payload for downstream agents
#   planning/rfp2_planning.json

Project Structure

web/
├── server.py                    — FastAPI backend (port 8000)
├── main.py                      — CLI entry point for batch processing
├── requirements.txt             — Python dependencies
├── package.json                 — Node.js dependencies
├── next.config.ts               — Next.js configuration
├── tsconfig.json                — TypeScript configuration
├── eslint.config.mjs            — ESLint configuration
├── postcss.config.mjs           — PostCSS/Tailwind configuration
│
├── rfp_agent/                   — Core RFP analysis engine
│   ├── __init__.py              — Package exports (RFPAnalysisAgent)
│   ├── agent.py                 — RFPAnalysisAgent class
│   ├── models.py                — Pydantic data models
│   ├── prompts.py               — Gemini prompt templates
│   └── document_loader.py       — PDF/DOCX/TXT loader with fallbacks
│
├── onboarding_agent/            — Google Drive → Sheets CV pipeline
│   ├── main.py                  — Pipeline orchestration
│   ├── config.py                — Environment configuration
│   ├── models.py                — CandidateProfile schema
│   ├── requirements.txt         — Additional Python deps
│   └── services/
│       ├── extractor.py         — Gemini CV extraction
│       └── google_api.py        — Google Drive/Sheets client
│
├── src/
│   ├── app/
│   │   ├── page.tsx             — Main UI (upload, analysis, matching)
│   │   ├── layout.tsx           — Root layout
│   │   ├── globals.css          — Global styles
│   │   └── api/
│   │       ├── process-rfq/     — RFP ingestion orchestrator
│   │       ├── matchmaking/     — Employee matching + PM selection
│   │       ├── outreach/        — ntfy.sh notification sender
│   │       └── employees/       — Employee CRUD + CV extraction
│   │           ├── route.ts     — GET all employees
│   │           ├── [id]/route.ts— POST upsert employee
│   │           └── extract/route.ts — POST CV upload + extraction
│   └── lib/
│       ├── gemini.ts            — Frontend → FastAPI bridge
│       └── storage.ts           — Local JSON persistence layer
│
├── data/
│   ├── employees.json           — Employee database
│   └── projects.json            — Project state store
│
└── public/
    └── cvs/                     — Uploaded CV file storage

Dependencies Reference

Python (requirements.txt)

Package	Version	Purpose
fastapi	≥ 0.111.0	HTTP API framework
uvicorn	≥ 0.29.0	ASGI server
python-multipart	≥ 0.0.9	File upload handling
google-genai	≥ 1.0.0	Google Gemini AI SDK
pydantic	≥ 2.0.0	Data validation and schemas
pdfplumber	≥ 0.10.0	PDF text extraction
python-docx	≥ 1.1.0	DOCX document parsing
python-dotenv	≥ 1.0.0	Environment variable loading
rich	≥ 13.0.0	Terminal formatting (CLI mode)
typer	≥ 0.12.0	CLI argument parsing

Python — Onboarding Agent (onboarding_agent/requirements.txt)

Package	Version	Purpose
google-generativeai	≥ 0.7.0	Gemini for CV extraction
google-auth	≥ 2.30.0	Google API authentication
google-auth-oauthlib	≥ 1.2.0	OAuth2 flow
google-api-python-client	≥ 2.130.0	Google Drive/Sheets SDK
cryptography	42.0.8	Auth token handling
pydantic	≥ 2.7.0	Schema validation
python-dotenv	≥ 1.0.1	Environment variables

Node.js (package.json)

Package	Version	Purpose
next	16.2.9	React framework (App Router)
react	19.2.4	UI library
react-dom	19.2.4	React DOM renderer
framer-motion	12.40.0	Animation library
@google/generative-ai	0.24.1	Gemini client (frontend utilities)
lodash	4.18.1	Utility functions
uuid	14.0.0	Unique ID generation
typescript	5.x	Type safety
tailwindcss	4.x	Utility-first CSS
eslint	9.x	Code linting

License

This project was built for the Tacto Hackathon 2025.

Name		Name	Last commit message	Last commit date
Latest commit History 35 Commits
data		data
onboarding_agent		onboarding_agent
public		public
rfp_agent		rfp_agent
src		src
.gitignore		.gitignore
AGENTS.md		AGENTS.md
CLAUDE.md		CLAUDE.md
README.md		README.md
README_OLD.md		README_OLD.md
architecture.md		architecture.md
eslint.config.mjs		eslint.config.mjs
main.py		main.py
next.config.ts		next.config.ts
package-lock.json		package-lock.json
package.json		package.json
postcss.config.mjs		postcss.config.mjs
requirements.txt		requirements.txt
server.py		server.py
tavily_test.py		tavily_test.py
tsconfig.json		tsconfig.json

Folders and files

Latest commit

History

Repository files navigation

Sales Engineering Agentic System

Table of Contents

Architecture

Pipeline Flow

Tech Stack

Prerequisites & Setup

Requirements

1. Clone & configure

2. Install dependencies

3. Run — open two terminals

macOS Troubleshooting

Environment Variables

Running the Application

Web Interface (recommended)

Demo Walkthrough

API Documentation

Python FastAPI Endpoints (port 8000)

POST /api/analyze

POST /api/analyze-multi

POST /api/extract-cv

POST /api/match

GET /health

Next.js API Routes (port 3000)

POST /api/process-rfq

POST /api/matchmaking

POST /api/outreach

GET /api/employees

POST /api/employees/[id]

POST /api/employees/extract

Core Modules

RFP Analysis Agent (rfp_agent/)

Matchmaking Agent

CV Extraction

Onboarding Agent (onboarding_agent/)

Outreach & Notifications

Data Models

RFPAnalysis (full output)

Planning Payload (slim handoff)

Project States

CandidateProfile / Employee

CLI Usage

Project Structure

Dependencies Reference

Python (requirements.txt)

Python — Onboarding Agent (onboarding_agent/requirements.txt)

Node.js (package.json)

License

About

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

`POST /api/analyze`

`POST /api/analyze-multi`

`POST /api/extract-cv`

`POST /api/match`

`GET /health`

`POST /api/process-rfq`

`POST /api/matchmaking`

`POST /api/outreach`

`GET /api/employees`

`POST /api/employees/[id]`

`POST /api/employees/extract`

RFP Analysis Agent (`rfp_agent/`)

Onboarding Agent (`onboarding_agent/`)

Packages