AI Project #3 · Agentic · Dual-model

Turn meetings into meaningful action.

Paste a transcript. A team of AI agents extracts decisions, action items, owners, deadlines and risks, cross-checks them against the source, and lets you ask questions in plain language.

Open the studio

Ingest

Paste a transcript from Zoom, Meet or Teams.

Extract

Agent 1 finds decisions, tasks, owners, deadlines and risks.

Verify

Agent 2 checks every item against the transcript line.

Plan and track

Agent 3 writes the summary and next steps. You approve.

Studio

Run the agent pipeline with your own key, or use the offline rule engine with no key at all.

Choose Run agent pipeline or Run offline engine.

AI assistant

Ask about the meeting you analyzed. Answers use only the transcript and cite line numbers.

Run the studio first, then ask me things like “What did Alex commit to?” or “Which items have no owner?”

How the agents work

Three specialists with one job each. If a model returns bad JSON, the agent retries once. If a provider fails, the other one takes over.

Extractor

Reads numbered transcript lines and returns structured JSON, including implicit tasks like “someone should look into this”.

Critic

Runs on the second model when you connect both keys. Removes items the transcript does not support, fixes owners and lowers confidence.

Planner

Writes the summary, the next steps and a ready-to-send follow-up email.

Model detection

Paste any key. “gsk_” keys go to Groq, “AIza” keys go to Gemini. The app lists the models on your account and picks the best current one.

Grounded assistant

Retrieves the most relevant transcript lines for each question, then answers with citations.

Human review

Items with no owner, no date or low confidence are flagged. Nothing is final until you tick it.

Technology

Static front end for free hosting. The same design scales to a service backend.

Groq API (Llama, GPT-OSS)Google Gemini APIMulti-agent orchestrationJSON-schema outputsRetrieval over transcript linesAuto model discoveryProvider failoverVanilla JS, zero build stepGitHub PagesResponsive, light and dark

transcript ─► Extractor (Groq) ─► Critic (Gemini) ─► Planner (Groq) ─► review board
                   │                    │                  │
                   └──── retries + provider failover ──────┘
question ─► line retrieval ─► Assistant (best available model) ─► cited answer

About the builder

Nikhil Chary Sriramoju

Final-year B.Tech student in Computer Science and Engineering. I build agentic AI systems, secure infrastructure and full-stack products, and ship them from my phone with Termux and GitHub Pages.

GitHub LinkedIn Email

Certifications and simulations

Forage: PwC Technology Consulting and Cyber Security, Datacom Cyber Security Operations and AI Accelerator, Clifford Chance, British Airways Data Science, Latham & Watkins, Deloitte Data Analytics and Technology
NASSCOM and Skill India: IT Forensic Analyst, Cloud Infrastructure Analyst, Endpoint Security, Blockchain Architect, Product Manager
Reliance Foundation: AI-ML Engineer Foundation, IoT Security Specialist, Industry 4.0 Robotics
NIELIT Yuva AI for All · Thiranex Data Analytics internship · Power BI (PHN Technology)

More projects

APEX-AIAgentic multimodal assistant with GraphRAG and MCP

BastionMesh AIMulti-agent mesh with consensus and CNN vision

SecureNav-NEXUSSecure autonomous robot and drone stack

PharmAi VisionMedicine identifier with CNN, ViT and OCR

Raft KV StoreDistributed key-value store with anomaly detection

VigilAI Engine MeshLLM serving with QoS gateway and observability

StreamGuardReal-time fraud detection pipeline

CareFlowHospital automation with AI chatbot