HAKASE — aKTIV8's AI Intelligence Engine

← LLM Index · Last updated: July 2026

What is HAKASE?

HAKASE is the AI intelligence layer built into aKTIV8. The name reflects its role as a domain expert ("hakase" means "doctor" or "expert" in Japanese). HAKASE is not a general-purpose chatbot — it is a purpose-built BD intelligence orchestrator that grounds all responses strictly in aKTIV8 platform data.

HAKASE can be accessed publicly at https://aktiv8.io/hakase. Full capabilities are available to authenticated platform users.

Core Capabilities

• Company intelligence Q&A: Ask HAKASE about any company in the platform — pipeline assets, recent funding, clinical milestones, regulatory history, investor relationships — and receive a grounded, sourced answer.
• Market analysis: Analyse therapeutic areas, competitive landscapes, deal activity patterns, and pipeline trends.
• BD opportunity identification: Find companies matching a set of criteria (stage, modality, geography, indication) that represent partnership or licensing targets.
• Asset-level analysis: Deep-dive into specific pipeline drugs or biologics — mechanism, clinical stage, regulatory pathway, competitive context.
• Outreach email drafting: Generate personalised BD outreach emails anchored to a specific company's real profile — no fabricated facts, all details drawn from platform data.
• Daily Brief generation: HAKASE writes the daily intelligence brief each morning, synthesising the previous day's signals into a structured narrative.
• BD report generation: Full-length reports including target assessments, opportunity scans, and market analyses — all grounded in corpus data.

Grounding Policy

HAKASE operates under a strict grounding policy: all outputs are anchored in facts present in the aKTIV8 data corpus. The system is designed to never fabricate company names, drug names, clinical results, funding amounts, or dates. If a fact is not in the platform data, HAKASE will say so rather than invent a plausible-sounding answer.

This is enforced through:

• RAG (Retrieval-Augmented Generation): relevant company profiles, news, and catalyst data are retrieved and injected into the prompt before generation
• Strip-fabricated-dates policy: date claims not supported by retrieved data are removed from responses
• Fail-closed subject lookup: if a company cannot be found in the corpus, the request fails gracefully rather than generating generic output

Multi-Model Architecture

HAKASE uses a multi-model routing system that dispatches different task types to the most appropriate underlying model:

• OpenAI GPT-4o / GPT-4o-mini: General Q&A, email drafting, conversational intelligence
• Anthropic Claude: Long-document analysis, report generation, structured output tasks
• Kimi (Moonshot AI): Tasks requiring extended context windows, especially for large corpus analysis

The routing system uses a static policy for initial dispatch and a scoreboard-based auto-routing system that adapts based on task performance data.

RAG Infrastructure

HAKASE's responses are grounded through a retrieval-augmented generation (RAG) layer:

• Document corpus: company profiles, news articles, catalyst events, clinical trial summaries
• Vector embeddings stored in PostgreSQL using pgvector
• Source registry tracking which data sources have been ingested
• New news articles are automatically ingested into the RAG corpus after each news scan
• Match feedback loop for improving retrieval quality over time

Daily Brief

Each morning at 7:00 AM Gulf time (3:00 AM UTC), HAKASE automatically generates the daily intelligence brief:

• Scans the previous day's signals (catalysts, funding, regulatory events, key news)
• Generates a structured narrative covering top themes
• Produces a concise headline summarising the day's themes (not company names)
• Brief is published at https://aktiv8.io/brief/[YYYY-MM-DD]
• Shareable via X (Twitter) and LinkedIn
• Delivered by email to subscribers

← Back to LLM Index