src/config.py

"""Configuration for Aethron Portfolio RAG — REVISED v3."""

import os

# Paths
BASE_DIR = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
DATA_DIR = os.path.join(BASE_DIR, "data")
INDEX_DIR = os.path.join(DATA_DIR, "index")
KB_PATH = os.path.join(DATA_DIR, "aragit_portfolio_kb.md")

# Embedding: BGE-small — 384-dim, ~50MB, strong technical retrieval
EMBEDDING_MODEL = "BAAI/bge-small-en-v1.5"
EMBEDDING_DIM = 384

# SLM: microsoft/Phi-3.5-mini-instruct via transformers
LLM_MODEL_ID = "microsoft/Phi-3.5-mini-instruct"
LLM_MAX_NEW_TOKENS = 512
LLM_TEMPERATURE = 0.1
LLM_DO_SAMPLE = False

# Retrieval
CHUNK_SIZE = 1024
CHUNK_OVERLAP = 200
TOP_K = 6
TOP_K_FETCH = 12

# Router keywords
RECRUITER_KEYWORDS = [
    "hire", "hiring", "salary", "available", "experience", "years", "skills",
    "kubernetes", "k8s", "docker", "python", "remote", "relocation", "visa",
    "full-time", "contract", "consulting", "rate", "contact", "email", "phone",
    "education", "degree", "university", "certification", "publication",
    "work", "worked", "job", "company", "career", "employed", "position",
    "kaggle", "medal", "competition", "master", "gold", "silver", "bronze"
]

TECHNICAL_KEYWORDS = [
    "architecture", "neuro-symbolic", "agentic", "rag", "mcp", "a2a",
    "temporal", "opa", "rego", "fhir", "ehr", "hipaa", "governance",
    "quantization", "lora", "qlora", "gemma", "mistral", "llm", "slm",
    "dag", "langgraph", "fastapi", "pydantic", "axiomis", "sentrixia",
    "nash", "equilibrium", "barnabus", "nayar", "zarif", "cogitator",
    "reflexa", "evolutio", "planning-rubicon", "world model", "repo",
    "repository", "github", "project", "implementation", "design",
    "stack", "layer", "pipeline", "orchestration", "clinical", "graphrag",
    "energy", "grid", "supply chain", "procurement", "finance", "regtech",
    "kyc", "aml", "biology", "protein", "molecular", "education", "tutoring",
    "marketing", "intent", "forecasting", "demand", "drift", "evaluator"
]

# Keyword → section_type fallback mapping
KEYWORD_SECTION_MAP = {
    "work": ["experience", "experience_summary_v2"],
    "worked": ["experience", "experience_summary_v2"],
    "job": ["experience", "experience_summary_v2"],
    "company": ["experience", "experience_summary_v2"],
    "career": ["experience", "experience_summary_v2"],
    "employed": ["experience", "experience_summary_v2"],
    "position": ["experience", "experience_summary_v2"],
    "skill": ["skills", "skills_summary"],
    "skills": ["skills", "skills_summary"],
    "expertise": ["skills", "skills_summary"],
    "technology": ["skills", "skills_summary"],
    "tech": ["skills", "skills_summary"],
    "publication": ["publications", "publications_summary"],
    "publications": ["publications", "publications_summary"],
    "paper": ["publications", "publications_summary"],
    "article": ["publications", "publications_summary"],
    "wrote": ["publications", "publications_summary"],
    "written": ["publications", "publications_summary"],
    "kaggle": ["kaggle_summary", "identity"],
    "medal": ["kaggle_summary"],
    "competition": ["kaggle_summary"],
    "master": ["kaggle_summary"],
    "gold": ["kaggle_summary"],
    "silver": ["kaggle_summary"],
    "bronze": ["kaggle_summary"],
    "clinical": ["clinical_ai_projects", "projects", "experience"],
    "healthcare": ["clinical_ai_projects", "projects", "experience"],
    "medical": ["clinical_ai_projects", "projects", "experience"],
    "marketing": ["marketing_ai_projects", "projects", "experience"],
    "advertising": ["marketing_ai_projects", "projects"],
    "supply chain": ["supply_chain_projects", "projects"],
    "logistics": ["supply_chain_projects", "projects"],
    "procurement": ["supply_chain_projects", "projects"],
    "energy": ["energy_projects", "projects"],
    "grid": ["energy_projects", "projects"],
    "biology": ["computational_biology_projects", "projects"],
    "protein": ["computational_biology_projects", "projects"],
    "molecular": ["computational_biology_projects", "projects"],
    "finance": ["finance_regtech_projects", "projects", "experience"],
    "regtech": ["finance_regtech_projects", "projects"],
    "kyc": ["finance_regtech_projects", "projects"],
    "aml": ["finance_regtech_projects", "projects"],
    "education": ["education_projects", "projects"],
    "tutoring": ["education_projects", "projects"],
    "neuro-symbolic": ["neuro_symbolic_projects", "architecture_deep_dive_axiomis", "projects"],
    "neuro symbolic": ["neuro_symbolic_projects", "architecture_deep_dive_axiomis", "projects"],
    "sota": ["projects", "projects_master_list", "neuro_symbolic_projects"],
    "github": ["projects", "projects_master_list"],
    "repo": ["projects", "projects_master_list"],
}

PERSONA_DEFAULT = "general"


def detect_persona(query: str) -> str:
    """Simple keyword-based persona detection."""
    q = query.lower()
    rec_score = sum(1 for kw in RECRUITER_KEYWORDS if kw in q)
    tech_score = sum(1 for kw in TECHNICAL_KEYWORDS if kw in q)
    
    if rec_score > tech_score and rec_score > 0:
        return "recruiter"
    elif tech_score > rec_score and tech_score > 0:
        return "technical"
    return PERSONA_DEFAULT


def get_boosted_sections(query: str) -> list:
    """Return section types to boost based on keyword fallback."""
    q = query.lower()
    boosted = set()
    for keyword, sections in KEYWORD_SECTION_MAP.items():
        if keyword in q:
            boosted.update(sections)
    return list(boosted)