Projects Done 2026

Production work completed this year. Use these as war stories in interviews.

---

HCLTech (2025–2026)

AskTGE — Enterprise AI Platform Performance Audit

Completed: Apr 2026 | Stack: Python FastAPI, LangGraph, AWS Bedrock (Claude Sonnet), DynamoDB, ECS, SSE
Impact: 13-issue perf audit across full stack

• Token cost: ~90% reduction via Bedrock prompt caching (cachePoint)
• SSE bandwidth: 50× reduction (O(N²) → O(N) delta streaming)
• Cold start: eliminated 700ms–1.5s ECS overhead via lazy init
• Parallel tool execution: asyncio.gather replaced sequential ToolNode

→ See full Q&A: [[Interview Prep/Coach & Tracking/Interview Questions/Past Experience Q&A For Interview]]
→ War story: [[Interview Prep/Coach & Tracking/Work Reference/AskTGE Work Summary]]

---

Chat History API — DynamoDB Schema + Encryption Redesign

Completed: Apr 2026 | Stack: Python FastAPI (async), DynamoDB Single-Table Design, Fernet, aioboto3
Impact: Zero-GSI schema with native sort ordering + race-condition-safe session promotion

• Native DynamoDB sort via epoch MS Sort Keys (no app-side sort)
• Atomic Bump Pattern: TransactWriteItems + ConditionExpression for ACID-safe concurrent writes
• KMS per-message round-trips eliminated via local Fernet encryption
• Zero-downtime dual-write migration plan (principal-architect-reviewed TDD v3)

→ See full Q&A: [[Interview Prep/Coach & Tracking/Interview Questions/Past Experience Q&A For Interview]]

---

TGE Logistics — Real-Time Driver Dispatch (WebSocket)

Completed: 2025 | Stack: React, Python FastAPI, MySQL, WebSockets (Aqchatter)
Impact: Sub-second real-time tracking for 250+ concurrent drivers

• Hybrid Hydration + Stream model (REST init + WS delta updates)
• O(1) key-value state vs O(N) array for driver reconciliation
• 500ms batch flush via useRef buffer — 70% CPU reduction
• React.memo per-pin isolation — 60FPS map rendering

→ See full Q&A: [[Interview Prep/Coach & Tracking/Interview Questions/Past Experience Q&A For Interview]]

---

Crest Data Systems / Splunk (2020–2024)

ML-Assisted Thresholding — Splunk ITSI

Completed: 2023 | Stack: React, Python/Django, Splunk Enterprise SDK, Scikit-learn
Impact: 90% search time reduction + Splunk.conf24 speaker

• Async job pattern for ML inference (non-blocking UI)
• Bulk processing: 10 KPIs per recommendation request
• Entity Adaptive Thresholding: nightly recalculation, dynamic policies
• Served 700+ global enterprise customers

→ See full Q&A: [[Interview Prep/Coach & Tracking/Interview Questions/Past Experience Q&A For Interview]]

---

Interview Positioning