Surya Kunapuli s3ak6i-dev

Full-Stack AI Developer · AI Agent Systems

About

Final year CS student shipping real products. I build full-stack AI applications end-to-end — backend APIs, data pipelines, ML integrations, and the frontend that makes it all usable.

Currently deep in AI agent systems and data infrastructure. Open to full-time roles and freelance projects building AI-powered applications.

Why s3ak6i? Inspired by the a16z naming convention — first letter, number of letters in between, last letter. Same idea, different name. Unusual on purpose.

Featured Projects

The intelligence layer beneath every AI system. Scans, cleans, enriches, and continuously improves the data that drives ML models.

Quality Engine with quantified model impact per issue
Pipeline Builder from plain English in under 2 minutes
Synthetic Data Engine + Active Learning Loop
Compliance Autopilot — GDPR, HIPAA, SOC 2

Visual canvas for orchestrating AI agents around a shared memory pool. Memory-first architecture — agents are workers, memory is the brain.

Drag-and-drop agent nodes with real-time WebSocket streaming
Conflict detection via cosine similarity + LLM contradiction check
D3 force graph memory dashboard
Canvas sharing with view/edit permissions

Real-time compliance guardrails, audit trails, and decision lineage for financial LLM agents. Wraps any agent framework automatically.

8 built-in rules mapped to GDPR, EU AI Act, MiFID II, Basel III
Hash-chain verification for tamper-evident audit trails
Human-in-the-loop approval for flagged decisions
SaaS control plane with live trace monitoring

Multi-tenant SaaS for product managers to organize and analyze user feedback. Surfaces underlying problems, not just feature requests.

Cluster signals into patterns automatically
Mandatory problem statements to enforce analytical thinking
Public share links for stakeholder access
First-class "Won't Solve" status for deliberate deprioritization

The credit rating agency for the agentic economy. MCP tells you how agents connect — AgentRank tells you whether to trust them.

Automated trust tiers (bronze → platinum) based on reputation
Full-text search with filtering by category and trust tier
Review system with star ratings and community voting
Side-by-side agent comparison tool

Open Source

Fixed a silent bug where agents using planning/reasoning would never proceed to task execution. Prompt templates instruct models to output READY or NOT READY, but the detection code only matched a legacy long-form phrase — causing a permanent mismatch. Added _is_ready() helper that handles both forms and added 12 regression tests.

PR #6261 · fix · crewAIInc/crewAI

Fixed a silent bug where tools with no title in their args_json_schema would crash at sandbox execution time. Two code paths generated the Pydantic class with different default names (DynamicModel vs Model), producing a NameError. Fix injects a consistent ToolArgs title at validation time in ToolCreate/ToolUpdate, with the helpers fallback aligned to match.

letta-ai/letta · fix · issue #3319