Asktra — The Cognitive Librarian for Software Systems

Inspiration

Software systems don’t just break because of bad code — they break because of vanishing context.

In modern enterprises, the “Truth” is fragmented across Slack messages, Jira tickets, Git commits, and release notes. When a senior engineer leaves, their why leaves with them. What remains is code without memory.

We were inspired to build Asktra after repeatedly watching teams spend 70% of their time playing forensic detective — digging through chaotic chat logs and stale documentation just to understand why a critical security setting was changed months ago.

We wanted to build a Cognitive Librarian that remembers why every line of code exists.


What It Does

Asktra is an investigative agent that reconstructs the Truth by weaving together fragmented signals from a company’s entire software stack.

It doesn’t just search — it reconciles.

Core Capabilities

  • Causal Reasoning
    Traces a production bug back to a “temporary hack” mentioned in Slack six months ago.

  • Contradiction Detection
    Flags when:

    • Code (Git) violates a security policy (Docs), or
    • A production release contradicts explicit human intent (Slack).
  • Sync Reality
    Generates PR-ready documentation and Reconciliation Bundles — including:

    • Post-mortem incident reports
    • Remediation code patches (diffs)
    • Stakeholder-ready Slack summaries

These bundles bridge the gap between intent and reality.


How We Built It

Asktra is powered by Gemini 3’s high-reasoning engine, using a multi-agent architecture:

The Agents

  • Context Weaver
    Leverages Gemini’s long-context window to ingest 50+ pages of policy and months of Slack history, performing temporal grounding to track how a crisis evolved over time.

  • Policy Enforcer
    Audits Git commits against SOC2 and internal compliance mandates, flagging violations even when Jira says an issue is “Resolved.”

  • Resource Allocator
    Reconciles Jira backlogs and human constraints to suggest “Third Path” solutions during organizational conflicts.

Deterministic Business Logic

We implemented a deterministic financial layer to compute Net Business Value:

We implemented a deterministic financial layer to compute Net Business Value:

Net Business Value = (Revenue Saved + Liability Mitigated) − Direct Cost

This ensures Asktra’s recommendations are not just correct — they are economically justified.

Frontend Design

The UI is built around Investigative Flow, allowing users to:

  • See the 22-step reasoning trace
  • Inspect every source
  • Verify exactly how Asktra reached its conclusions

Challenges We Ran Into

Causal Disconnect

One of the hardest problems was Causal Disconnect — when Slack, Jira, Git, and Docs all disagree.

We had to teach Gemini not to choose a source, but to:

  • Identify the Truth Gap
  • Flag it as an active risk
  • Preserve contradictions instead of smoothing them over

High-Stakes Security

Because Asktra operates in security-critical environments, we implemented a cryptographic Audit Signature (SHA-256) for every reasoning trail. This ensures Asktra’s outputs are tamper-evident and audit-admissible in real-world compliance reviews.


Accomplishments We’re Proud Of

The Reconciliation Bundle is our biggest win.

Watching Asktra:

  • Detect a 90-second timeout violation,
  • Identify the human intent behind it,
  • Generate the exact code diff to fix it,
  • Write the post-mortem,
  • And draft the Slack message to align the team…

…was a true wow moment.

We also proved that AI can handle Triple Conflict scenarios — balancing Engineering, Security, and Business constraints simultaneously.


What We Learned

We learned that context is the ultimate currency.

Modern reasoning models like Gemini 3 aren’t just conversational — they are capable of organizational logic across time, policy, and human intent.

We also learned the importance of Human-in-the-Loop design. Asktra doesn’t replace engineers; it acts as an Investigative Co-pilot, presenting evidence and reasoning for human authorization.


What’s Next for Asktra

Our next milestone is Proactive Drift Detection.

Instead of waiting for someone to ask a question, Asktra will:

  • Monitor conversations and commits in real time
  • Detect when decisions begin to drift away from policy
  • Alert teams before a security or compliance failure happens

Our goal is for Asktra to become the immune system for the world’s most critical software.


Built With

  • Gemini 3
  • Multi-Agent Systems
  • Vertex AI
  • Node.js / React
  • SHA-256 Cryptographic Auditing

Built With

  • causal-reasoning
  • custom-css
  • deterministic-business-logic
  • eslint
  • fastapi
  • gemini-3
  • google-gemini-api
  • google-genai
  • google/genai
  • javascript
  • multi-agent-systems
  • net-business-value
  • next.js-14
  • node.js
  • npm
  • pydantic
  • python
  • react-18
  • react-markdown
  • sha-256-audit-signatures
  • soc2-compliance
  • typescript
  • uvicorn
  • vercel
  • vite
Share this project:

Updates