Thinking steps stage
Thinking steps and answer stage
Answer page
Answer page
Source-aware
New document creation for future use
Sample data used
Reconciliation Bundle

Asktra — The Cognitive Librarian for Software Systems

Inspiration

Software systems rarely fail because of bad code alone — they fail because of lost context.

In modern engineering teams, intent is scattered across Slack threads, Jira tickets, Git commits, and policy documents. When engineers leave, the why behind decisions disappears. What remains is code without memory.

Teams often spend hours reconstructing decisions just to understand why a production change was made months ago. Asktra was built to eliminate that forensic effort.

What It Does

Asktra is an investigative AI agent that connects Slack, Jira, Git, and internal policies to reconstruct the reasoning behind code changes.

It does not just summarize — it reconciles intent with reality.

Core Capabilities

Causal Traceability
Links a production issue to the original Slack discussion, Jira task, and related commit.
Policy Violation Detection
Flags mismatches between code and compliance requirements (e.g., missing approvals, security policy drift).
Reconciliation Bundles
Automatically generates:
- Post-mortem reports
- Suggested remediation code diffs
- Stakeholder-ready Slack summaries

Each bundle connects decision → action → impact in a verifiable chain.

How We Built It

Asktra uses Gemini’s long-context reasoning within a multi-agent architecture.

System Architecture

Context Weaver
Ingests Slack threads, Jira tickets, Git commits, and policy documents to build a temporal reasoning graph.
Policy Enforcer
Audits code changes against SOC2 and internal compliance requirements.
Conflict Detector
Identifies contradictions between documented intent and deployed reality.

Deterministic Audit Layer

Each reasoning output is hashed using SHA-256 to generate a tamper-evident audit signature.

We also compute:

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

This ensures recommendations are technically valid and economically justified.

Challenges We Solved

Cross-System Contradictions

Slack, Jira, and Git frequently disagree.
Instead of forcing alignment, Asktra preserves contradictions, flags them as risk, and presents traceable evidence.

High-Stakes Environments

Security and compliance require verifiable outputs.
We implemented cryptographic audit signatures to ensure reasoning trails are transparent and reviewable.

Accomplishments

In testing, Asktra successfully:

Detected a timeout policy violation in production code
Identified the Slack discussion that justified the temporary change
Generated a compliant code patch
Drafted the post-mortem
Prepared a stakeholder summary

This demonstrated true end-to-end reconciliation.

What We Learned

Context is infrastructure.

Modern reasoning models can connect information across time and systems — but trust requires transparency. Human-in-the-loop review remains essential.

Asktra acts as an investigative co-pilot, not an autonomous decision-maker.

What’s Next

We are building Proactive Drift Detection, where Asktra monitors conversations and commits in real time and alerts teams before policy deviations reach production.

Our vision is to make Asktra the memory and governance layer for modern software teams.