COMPLETE Hackathon Submission - All requirements fully implemented and verifiable.
A production-grade AI-powered SRE agent that automatically triages, analyzes, and routes production incidents from submission through resolution.
Submission by Team: Agentlemen
The agent ingests multimodal incident reports (text + images/logs), performs intelligent analysis using vector similarity and LLM reasoning, creates Jira tickets, notifies teams, and tracks the full lifecycle through resolution notifications.
Hackathon Submission: COMPLETE full-stack agent addressing core E2E flow + ALL minimum requirements (multimodal input, guardrails, observability, integrations, e-commerce codebase). No mocked components - all integrations are functional for demo purposes. Only some mocked data.
Full step-by-step instructions in QUICKGUIDE.md.
Prerequisites: Docker & Docker Compose (v20.10+), Git, and a Google AI API key.
# 1. Clone
git clone git@github.com:AlasAltum/Agentleman-Hackathon-Softserve.git
cd Agentleman-Hackathon-Softserve
# 2. Configure — fill in the 6 required values
cp .env.example .envRequired values in .env:
POSTGRES_PASSWORD=your_password_here
GOOGLE_API_KEY=your_google_api_key_here # https://aistudio.google.com/apikey
ATLASSIAN_EMAIL=your_jira_email@example.com
ATLASSIAN_API_TOKEN=your_jira_api_token # https://id.atlassian.com/manage-profile/security/api-tokens
NYLAS_API_KEY=your_nylas_api_key # https://www.nylas.com/
NYLAS_GRANT_ID=your_nylas_grant_id
COHERE_API_KEY=your_cohere_api_key # https://cohere.com/ (optional — enables semantic reranking)# 3. Start everything
docker compose up --build
# 4. Verify
curl http://localhost:8000/health
# 5. Submit a test incident
curl -X POST http://localhost:8000/api/ingest \
-F "text_desc=Production database experiencing connection timeouts since 14:10 UTC" \
-F "reporter_email=engineer@example.com"No local Python, Node.js, or database setup needed — everything runs in Docker.
| Requirement | Implementation | Evidence & Verification |
|---|---|---|
| Multimodal Input | Text + image/log file support via /api/ingest with file attachment handling |
backend/src/api/routes/incident_routes.py accepts file_attachments (PDFs, images, logs); content extracted via LlamaHub. Verify: Submit incident with file attachment → check logs for extraction |
| Guardrails | Input validation, prompt injection defense, content threat detection, LLM relevance check | backend/src/guardrails/ module: validators.py (MIME), input_guardrails.py (threats), relevance_guardrail.py (LLM check). Verify: Submit malicious input → rejected with error |
| Observability | Structured logs (structlog), end-to-end traces (MLflow autolog), Prometheus metrics, Loki aggregation, Grafana dashboards | Full stack: MLflow (traces), Prometheus (metrics), Loki (logs), Grafana (dashboards). Verify: Submit incident → check http://localhost:5001 (traces), http://localhost:3000 (dashboards) |
| Integrations | Jira ticketing plus Nylas email notifications for team and reporter flows | backend/src/services/jira/, backend/src/services/notifications/. Verify: Submit incident → Jira ticket created, team email sent, reporter email sent |
| E-Commerce Codebase | Medusa e-commerce backend + Next.js storefront (medium complexity) | ecommerce-platform/ with shared PostgreSQL. Verify: Visit http://localhost:8001 → click "Having problems?" → submit incident |
1. SUBMIT via UI or API
├─ Option A: SRE Incident Report Platform (http://localhost:8000)
├─ Option B: E-Commerce Storefront "Having problems?" button (http://localhost:8001)
└─ Option C: RESTful API (/api/ingest)
↓
2. GUARDRAILS VALIDATE (MIME + magic bytes + threat patterns + LLM relevance checks)
↓
3. MULTIMODAL EXTRACTION (PDFs/images/logs processed)
↓
4. AI TRIAGE (vector search + reranking + LLM classification)
↓
5. PARALLEL TOOL ANALYSIS (business_impact, codebase_analyzer, telemetry_analyzer)
↓
6. TICKET CREATED in Jira (with LLM-generated summary and full context)
↓
7. TEAM + REPORTER NOTIFIED (email via Nylas)
↓
8. ENGINEER RESOLVES in Jira
↓
9. WEBHOOK triggers REPORTER RESOLUTION NOTIFICATION + KB upsert to Qdrant
Verification:
# Submit test incident
curl -X POST http://localhost:8000/api/ingest \
-F "text_desc=Production database timeout" \
-F "reporter_email=test@example.com"
# Check workflow execution
docker logs hackaton-backend | grep -A 20 "ingest_started"
# Check observability
open http://localhost:5001 # MLflow traces
open http://localhost:3000 # Grafana dashboards (admin/admin)- ✅ Severity scoring (via guardrails + LLM classification)
- ✅ Historical incident deduplication (via vector similarity + reranking)
- ✅ Multi-agent coordination (Routing + tool dispatch)
Backend Services:
- API Framework: FastAPI (async Python web framework)
- Workflow Engine: LlamaIndex Workflows (event-driven orchestration)
- Vector Database: Qdrant (semantic search for incident similarity)
- Reranking: Cohere Rerank API
- Logging: structlog (structured, request-correlated logs)
Frontend Services:
- SRE Platform: Custom FastAPI-served HTML forms
- E-Commerce Storefront: Next.js + Medusa (headless commerce)
- Shared Database: PostgreSQL (customer + incident data)
Observability Stack (All Open Source):
- Metrics: Prometheus (time-series metrics collection)
- Logs: Loki (centralized log aggregation)
- Traces: MLflow (end-to-end workflow tracing)
- Dashboards: Grafana (visualization platform)
Infrastructure:
- Containerization: Docker + Docker Compose
| Service | Port | Description | Type |
|---|---|---|---|
| Backend API | 8000 | FastAPI incident ingestion + webhooks | Python/FastAPI |
| PostgreSQL | 5432 | Incident metadata + ML state (open source) | Database |
| Qdrant | 6333 | Vector DB for semantic search (open source) | Vector DB |
| Medusa API | 9000 | E-commerce backend (open source) | Node.js |
| Next.js Storefront | 8001 | E-commerce UI (open source) | React/Next.js |
| Grafana | 3000 | Dashboards (open source) | Visualization |
| Prometheus | 9090 | Metrics collection (open source) | Metrics |
| Loki | 3100 | Log aggregation (open source) | Logs |
| MLflow | 5001 | Traces & experiment tracking (open source) | Tracing |
Three validation layers on all incident inputs:
- MIME Type Validation — Allows: PNG, JPG, GIF, WEBP, TXT, LOG, CSV | Blocks: Executables, archives, scripts
- Pattern-Based Threat Detection — Flags SQL injection attempts, shell commands, XSS, path traversal via pattern matching
- LLM Relevance Check — Uses LLM to verify incident is SRE-related; rejects off-topic or adversarial inputs
Code references: guardrails/validators.py, guardrails/input_guardrails.py, guardrails/relevance_guardrail.py. Covered by integration tests in backend/tests/test_ingest_integration.py.
Supported file types: PNG, JPG, GIF, WEBP, TXT, LOG, CSV (extracted via Gemini vision + text parsing).
Example (API):
curl -X POST http://localhost:8000/api/ingest \
-F "text_desc=API degradation" \
-F "reporter_email=john@acme.com" \
-F "file_attachments=@error.log" \
-F "file_attachments=@screenshot.png"View from MLFlow
View from Grafana
Jira Cloud API (backend/src/integrations/jira/)
- ✅ Creates issues on incident ingestion with LLM-generated summary
- ✅ Listens for Jira webhooks on issue status change (+ local polling fallback)
- ✅ Updates reporter email on resolution
Nylas Email Notifications (backend/src/services/notifications/)
- ✅ Sends email alerts to the engineering team
- ✅ Sends acknowledgement email to reporter on ticket creation
- ✅ Sends resolution email to reporter when Jira marks the issue as resolved
This submission uses Google Gemini as the LLM and embedding provider:
| Config | Value |
|---|---|
LLM_PROVIDER |
google |
LLM_MODEL |
gemini-2.5-flash |
EMBED_PROVIDER |
google |
EMBED_MODEL |
gemini-embedding-2-preview |
A free API key is available at https://aistudio.google.com/apikey — the only LLM credential required.
# Complete stack boots with one command
docker compose up --build- ✅ README.md — Architecture overview, setup, compliance with hackathon requirements
- ✅ QUICKGUIDE.md — 5-minute setup guide
- ✅ AGENTS_USE.md — Agent capabilities, use cases, security, observability evidence
- ✅ SCALING.md — Production deployment & scaling strategies
- ✅ docker-compose.yml — Single-command build & run
- ✅ .env.example — All required variables with placeholders
- ✅ LICENSE — MIT license for open source
- ✅ Dockerfile(s) — Backend, frontend containerization
- Code Quality: All Python files are syntactically correct and importable
- Configuration:
.env.examplecontains all required variables with realistic defaults - Dependencies:
pyproject.toml,package.jsonfiles are complete with all packages listed - Docker: All services have working Dockerfiles and health checks
- Tests: Basic test structure exists in
backend/tests/(pytest runnable)
For logs and debugging:
docker compose logs hackaton-backend | tail -50
docker compose logs -fSpecific issues:
- API 500 Error: Check
.envfor valid LLM_PROVIDER and API key - Qdrant not responding:
docker compose restart hackaton-qdrant - Database error:
docker compose exec db psql -U postgres -c "SELECT version();" - MLflow traces missing: Verify MLflow is running at http://localhost:5001
- QUICKGUIDE.md — 5-minute setup guide
- backend/README.md — Backend architecture details
- ecommerce-platform/README.md — Storefront setup
- observability/README.md — Monitoring & tracing
- SCALING.md — Production deployment & scaling strategies
- AGENTS_USE.md — Agent capabilities, use cases, security measures