# Enterprise Brain Scrum - March 29, 2026 **Date:** March 29, 2026 **Time:** 19:58 PDT **Duration:** ~27 minutes **Participants:** Rajashekar G (Raj), Yeshwanth Reddy Yerraguntla (Yesh), Phawhan Saii Gajjalakonda (Pawan), Vara Kumar Jagarapu (Vara) **Source:** [Enterprise Brain Scrum - 2026_03_29 19_58 PDT - Notes by Gemini.txt](Enterprise%20Brain%20Scrum%20-%202026_03_29%2019_58%20PDT%20-%20Notes%20by%20Gemini.txt) --- ## Act I - API Key Consumption & Cost Tracking (~00:00 - ~05:29) The meeting opened with weekend greetings and a brief exchange about the handbook. Yesh mentioned he's still figuring out how to enforce handbook practices on intermediate repos but expects a solution by next week. He highlighted a recent win: the Jira agent can now directly communicate with Jira. The conversation quickly turned to API key usage and cost concerns. Raj flagged that usage jumped from $80 on Friday night to $84.50 — meaning $4.50 was consumed over the weekend. Yesh clarified that he stopped using the shared key over a week ago and switched to local LLM for his work. Looking at the usage dashboard, they identified that the March 29th spike came from third-row embedding consumption totaling $4.40. Yesh questioned why the OptiNova team would be using the Enterprise Brain key for load testing or other work. Raj committed to checking with Rahul, Manisha, and Prabhakar to understand the usage. > _"First thing is you have to invalidate all the old keys and then give one key to every person because without invalidation uh you can't enforce"_ > — Yeshwanth Reddy Yerraguntla _(~05:29)_ Yesh stressed the urgency of per-user key tracking. He pointed out that without knowing who used what, they have no accountability. The team agreed that user-level tracking is 100% better than project-level tracking for visibility and control. ## Act II - Frontend Design Updates & Plotly Integration (~06:45 - ~08:52) Raj shared that the design team delivered new mockups on Friday. Over the weekend, Amulia implemented the landing screen in the Enterprise Brain frontend folder, including backend changes to communicate over port 8010. Harsh built visualization components using Plotly based on the design specs, complete with mock data. The UI components exist, but there's a critical gap: > _"I see that LLM is still not talking to it."_ > — Yeshwanth Reddy Yerraguntla _(~07:40)_ Right now, the chart components are pure frontend with no agent communication layer. The team needs to figure out how to inject data from the agent into these Plotly visualizations. Raj marked this as the **first priority** and committed to conquering it by **Wednesday's delivery deadline**. Raj mentioned he couldn't get the frontend running locally yet and will sync with Amulia once they're in the office. ## Act III - Data Understanding & Schema Design (~08:52 - ~13:04) Pawan joined and provided an update on the data work. The team is focusing on the commercial management early warnings sheet from the Tata Steel dataset. Pawan used Claude to understand the documents and distinguish between raw data sheets and derived report sheets. He generated SQL table schemas with proper column definitions and business context. The schema includes application-level context explaining that the database supports commercial and contract management functions for a major industrial construction project governed by NC4 standards. > _"So we have to insert the data into these tables and then we have to check to NL scale and check uh and uh and then but these this schema is not giving any uh information to the agent um with user perspective"_ > — Phawhan Saii Gajjalakonda _(~10:02)_ Pawan recognized that raw schemas lack business meaning, so he asked Claude to generate column-level comments explaining what each field represents from a user perspective. Now the schema includes both DDL scripts and contextual metadata. Yesh reviewed the application context and confirmed the language was clear enough for an LLM to understand NC4 compliance requirements and early warning notifications. ## Act IV - Use Case Scoping & Project Strategy (~13:04 - ~18:26) Vara provided strategic context from recent client discussions. The team identified five potential use case areas across the Tata Steel dataset but decided to focus on **one area initially**: contract administration, specifically early warnings. The data is organized into two categories: 1. **Data source tables** — raw data with plain column structures 2. **Report sheets** — derived insights with pivot tables and dashboard-style layouts > _"So what we thought of doing like also goal are included there and we just want to dump this first of plain tables into SQL with the same names like as we have in the uh column names right considering as a one to one mapping to the database"_ > — Vara Kumar Jagarapu _(~14:15)_ The strategy is to first do a 1:1 mapping of raw sheets into SQL tables, then derive instructions and test questions from the report sheets. The team is **not** trying to deliver all five use cases at once — instead, they'll demonstrate depth in one area first before expanding. Vara shared documents with Pawan and Raj. Yesh asked whether the data contained anything sensitive like pricing. Vara confirmed there's nothing sensitive in the dataset. Yesh noted the team is still using Google Docs because they're leveraging Gemini directly for some exploration work, but emphasized that finalized materials should move to permanent documentation. ## Act V - Implementation Approach & Tooling (~18:26 - ~27:53) Yesh provided critical implementation guidance to avoid wasting time on premature integration: > _"Pan what I suggest is uh uh don't try to integrate directly into enterprise brain right now. Generate the SQL. Okay, you generate the DB, you give the full schema to uh what's it uh to some copilot in VS code only."_ > — Yeshwanth Reddy Yerraguntla _(~19:03)_ The workflow Yesh recommended: 1. Generate the SQL DB and full schema 2. Attach the schema file to a VS Code Copilot conversation 3. Give Copilot a raw function to connect to SQL and execute queries (no hardcoded credentials in the prompt) 4. Simulate NLP-to-SQL in Copilot to validate the prompts work correctly 5. Use Copilot itself to refine the schema and prompts 6. **Only after validation**, integrate into Enterprise Brain This approach isolates SQL prompt quality from integration complexity. If something doesn't work, the team will know immediately whether it's a schema issue, a prompt issue, or an integration issue. Raj asked whether Pawan should use SolvIt instead of Copilot for data exploration. Yesh confirmed SolvIt will be faster for this kind of work — especially when uploading Excel files, DB schemas, and iterating on business understanding. Yesh enabled SolvIt access and helped Raj create a Tata Steel folder for uploading project files. The team agreed that Pawan and Raj should use the documents Vara shared to derive test questions from the report sheets, since those capture the actual insights and queries users care about. ## Todos Check with OptiNova team (Rahul, Manisha, Prabhakar) about March 29 API key usage ($4.40 third-row embedding)
👤 Rajashekar G 📅 ASAP Invalidate all old API keys and generate per-user keys for tracking and enforcement
👤 Rajashekar G / Yeshwanth Reddy Yerraguntla 📅 Next week Debug frontend setup not running locally — sync with Amulia in office
👤 Rajashekar G 📅 Monday (in office) Implement agent communication layer to inject data into Plotly chart components (first priority)
👤 Rajashekar G / Harsh 📅 Wednesday delivery Get status update from Ablash on NLP to SQL agent setup
👤 Rajashekar G 📅 Monday Generate SQL DB with 1:1 table mapping, insert data, and test schema with business context
👤 Phawhan Saii Gajjalakonda 📅 This week Test NLP-to-SQL prompts in VS Code Copilot with schema attachment and DB connection function before integrating into Enterprise Brain
👤 Phawhan Saii Gajjalakonda 📅 This week Derive test questions from report sheets to validate NLP-to-SQL implementation
👤 Phawhan Saii Gajjalakonda / Rajashekar G 📅 This week - [ ] Upload Excel files, DB schema, and business context documents to SolvIt Tata Steel folder for data exploration | author: Phawhan Saii Gajjalakonda | deadline: This week