---

Production 5-agent pipeline that automates the full pre-sales cycle — from discovery call notes to client-ready proposal. Built to replace 3–4 hours of manual proposal writing per SISA engagement.

Architecture

DiscoveryAgent → ScopingAgent → PricingAgent → ProposalAgent
                                       ↑
                                  CriticAgent (adversarial review)

Agent	Role
DiscoveryAgent	Extracts BANT from discovery notes — budget, authority, timeline, trigger event
ScopingAgent	Maps requirements to SISA service lines (6 BUs, 22 services)
PricingAgent	Estimates effort and builds commercial structure
CriticAgent	Adversarial pass — challenges assumptions, flags scope creep risks before proposal
ProposalAgent	Synthesises all upstream output into client-facing proposal

The CriticAgent runs between Pricing and Proposal — it's not a safety net, it's a mandatory commercial review that catches over-promises before they're committed.

Service coverage

Pen Testing: network · web app · mobile · API · cloud · red team
Compliance: PCI DSS v4.0 · ISO/IEC 27001 · SOC 2 · HIPAA
DFIR: IR retainer · forensics · tabletop · threat hunting
VA&M: full vulnerability management lifecycle

Stack

• AI: Groq — llama-3.3-70b-versatile (open source, <1s per agent call)
• Orchestration: custom multi-agent pipeline — no framework overhead
• UI: Streamlit

Setup

git clone https://github.com/sanjayrkshetty/proposal-engine
cd proposal-engine
pip install -r requirements.txt
cp .env.example .env   # add GROQ_API_KEY
streamlit run app.py

Why Groq + Llama over OpenAI

Each agent call is <1s on Groq. A full 5-agent pipeline completes in ~4s — interactive, not a background job. Llama 3.3 70b matches GPT-4 class quality on structured extraction tasks at a fraction of the cost.

---

Part of @sanjayrkshetty's AI security portfolio