Enterprise AI · Strategy · LangGraph

AI Use Case
Scoper

"8 guided questions. Share your tech audit doc if you have one — we'll pre-fill the technical context. Come back in 90 days: the plan evolves with you."

View on GitHub Read the Docs

$ python main.py scope --interactive

# Do you have a tech audit doc to share? [y/N]
# > y  →  tech_audit.pdf (42 chunks)
# Graph RAG: 47 entities · pre-filling 5 fields
# [3/8] AI ambition? → always asked

readiness_score: 62 / 100
recommended: Internal Doc Q&A
session saved · id: a3f9…

# 90 days later…
$ python main.py checkin a3f9… --interactive

# LangGraph: 6 nodes · delta extracted
readiness_score: 75 / 100
delta: +13
evolution: Doc Q&A shipped. ML hire
unlocked contract review.

Problem

Every enterprise AI engagement starts the same way: leadership wants AI, nobody agrees on what for, and the first project gets chosen based on whoever makes the loudest argument in a workshop. Six months and a failed pilot later, the appetite for AI has halved.

AI Use Case Scoper runs 8 guided questions to understand what a company wants from AI. Share a tech audit doc if you have one — Graph RAG pre-fills the technical context so you spend the conversation on what matters: AI ambition, pain points, and priorities. The output is a ranked use case list with a recommended first project and a 90-day plan. Come back three months later: LangGraph compares reality against the roadmap and generates an evolved plan with an explicit delta. No starting from scratch.

Architecture

Questions first.
Docs welcome.

v2 adds hybrid document enrichment, company memory, and a LangGraph check-in pipeline — so the plan evolves as the company does.

Ingest

Optional first step: share a PDF, DOCX, or CSV. Graph RAG (NetworkX + ChromaDB) extracts systems, teams, compliance, and tech stack — pre-filling technical fields where confidence ≥ 70%

Score

Profile assembled with per-field confidence scores. Low-confidence fields surfaced for human review. Pre-scorer applies rule-based readiness modifiers

Scope

Claude applies industry heuristics to generate and rank use cases by feasibility, ROI, and risk. Session persisted to Postgres with company memory

Evolve

90 days later: LangGraph pipeline compares check-in notes against the original roadmap, re-scores use cases, and generates an evolved plan with explicit delta

Features

Honest.
Actionable.

Built for conversations with non-technical stakeholders. Every output is specific enough to put in a board deck — and specific enough to revisit in 90 days.

Hybrid Document Enrichment

At the start of the 8-question flow, the CLI optionally asks: "Do you have a tech audit doc?" If yes, Graph RAG (NetworkX + ChromaDB) extracts technical context from the PDF/DOCX and pre-fills fields where confidence ≥ 70%. Qualitative questions — AI ambition, pain points, timeline — are always asked. Docs enrich; they don't replace the conversation.

LangGraph Check-in Pipeline

Come back 90 days later. A stateful LangGraph pipeline compares your check-in notes against the original roadmap: re-scores use cases, shifts milestones, and generates an evolved plan with an explicit delta summary.

Company Memory

Every scoping session is persisted. Previous readiness scores, recommended projects, and resolved blockers are injected as context into future sessions — so recommendations compound over time, not repeat.

Use Case Prioritisation

Every use case scored on feasibility, ROI, and risk — ordered by weighted priority score. Output is readable by product leads and CTOs without a data scientist to interpret it.

Industry-Specific Heuristics

Legal and finance → document analysis. E-commerce → support automation. Ad-tech → creative generation and campaign diagnosis. Recommendations are grounded in what actually ships in each vertical.

Jira / ClickUp Export

The 90-day roadmap exports directly to Jira (epics + stories) or ClickUp (tasks with phase tags) via REST — no SDK dependency, no bidirectional sync overhead.

Quickstart

Running in
three minutes.

Setup

git clone https://github.com/swapnanil/ai-use-case-scoper
cd ai-use-case-scoper
cp .env.example .env   # add your ANTHROPIC_API_KEY
docker-compose up api  # starts API + Postgres

interactive hybrid mode — 8 questions + optional doc enrichment

docker-compose run --rm -it cli scope --interactive

skip the doc prompt — pure 8-question flow

docker-compose run --rm -it cli scope --interactive --no-ingest

v2 — run a check-in 90 days later (LangGraph pipeline)

python main.py checkin <company-id> \
  --session-id <session-id> --interactive

v2 — export roadmap to Jira

curl -X POST http://localhost:8000/sessions/<id>/export/jira \
  -d '{"jira_base_url": "...", "project_key": "AI", "api_token": "..."}'

Example

Questions in.
Evolved plan out.

Initial scope — hybrid 8-question flow

# Do you have tech docs to share? y
# Graph RAG: 47 entities extracted
pre_filled:
  tech_stack, compliance_requirements,
  has_ml_engineers, engineering_team_size
always_asked:
  ai_ambition, pain_points, timeline

# POST /scope → scoping run
readiness_score: 62
recommended:
  "Automated Report Generator"
session_id: "a3f9c…"

90 days later — evolved plan (LangGraph)

# POST /companies/:id/checkin
readiness_score:       74
readiness_score_delta: +12

evolution_summary
"Report generator shipped Week 4.
 Data blocker resolved — Q&A
 agent now feasible."

milestone_shifts:
  "Week 3–6 removed (shipped)"
dropped_use_cases: []

AI Use CaseScoper

Questions first.Docs welcome.

Honest.Actionable.

Running inthree minutes.

Questions in.Evolved plan out.

Five more tools.Same standard.

AI Use Case
Scoper

Questions first.
Docs welcome.

Honest.
Actionable.

Running in
three minutes.

Questions in.
Evolved plan out.

Five more tools.
Same standard.