Software Engineer
Aditya Kalambe
I ship things that work in production – not just in demos.
Summary
Software Engineer shipping production-grade agentic AI systems, not just prototypes. Strong backend fundamentals in Python and Node.js, hands-on with LangGraph, LangChain, RAG pipelines, and LLM-as-judge evaluation, end to end from model integration to deployment.
Skills
Programming Languages
Python, JavaScript, C++, SQL
AI & LLM Engineering
Agentic AI, LangGraph, LangChain, RAG Pipelines, Vector Databases (FAISS, Pinecone, Chroma), Knowledge Graphs, LLM Integration, LLM-as-judge Evaluation, Prompt Engineering, Voice AI Systems
Backend Development
Node.js, Express.js, FastAPI, Django, Django REST Framework, REST APIs, WebSockets
Frontend Development
Next.js, React.js
Database Management
PostgreSQL, MongoDB, MySQL, Redis
Cloud & DevOps
AWS (EC2, S3, Lambda), Docker, Kubernetes, Git
Development Tools
Prisma ORM, BullMQ, Postman, Jupyter Notebook
Core Concepts
Data Structures & Algorithms, System Design, Data Cleaning & Deduplication, API Security, Performance Optimization, Unit Testing, Integration Testing
Work Experience
Forward Deployed Engineer
Chat360
- Replaced a manual spreadsheet-driven WhatsApp workflow with a production React + FastAPI dashboard, automating a 12-step service journey with webhook-driven messaging.
- Migrated the client's system of record from Google Sheets to Supabase, cleaning and de-duplicating messy source data, and built CX instrumentation (CSAT, engagement index, WhatsApp join rates).
- Built and hardened a bilingual (English/Hindi) AI voicebot for FMCG product-complaint intake, designing a 40+ category classification system and resolving 15+ production-blocking defects via transcript-based QA.
Software Engineer
500X Tech Labs
- Built a full-stack HRMS platform with Next.js, Express.js, Prisma, and PostgreSQL, delivering employee management and workflow automation features.
- Optimized REST APIs, improving response consistency and query latency through efficient design.
- Diagnosed and resolved production issues in API routing, database relationships, data integrity, and deployments, improving platform stability.
AI Intern
Kuhoo Finance
- Led development of a real-time agentic voice assistant integrated with telephony systems, orchestrating multi-step conversational workflows with LangGraph and LangChain.
- Built a low-latency bidirectional audio pipeline using WebSockets, reducing caller wait times by 30%.
- Achieved 85% transaction completion through retry mechanisms, monitoring, and performance-driven workflow optimization.
- Built a Knowledge Graph-powered discussion assistant using FAISS-backed RAG, LLM refinement, and Monte Carlo Tree Search for context-aware recommendations, evaluated with an LLM-as-judge pipeline.
Projects
OfflinePDF
TypeScript, Next.js, Web Workers, WebAssembly (qpdf-wasm), pdf-lib, pdf.js, Tesseract.js, PeerJS
- Built a privacy-first suite of 42 PDF tools running entirely in-browser with zero server uploads, offloading processing to an isolated background Web Worker with on-demand engine loading to handle files up to 300 MB without main thread blocking.
- Implemented client-side AES-256 encryption/decryption via WebAssembly (qpdf-wasm), on-device OCR using bundled Tesseract models, and direct browser-to-browser WebRTC data channels (PeerJS) for P2P file transfers.
- Extracted core utilities into a published standalone npm package (offlinepdf-sdk) and built a local MCP server (offlinepdf-mcp) enabling Claude Desktop and Claude Code to execute 17 document tools completely offline.
SRE Agent
Node.js, Express, Next.js, Groq, Postgres (Supabase), Render, Vercel, Prometheus, GitHub Actions
- Built a self-healing distributed system: three e-commerce microservices plus a control-plane agent that polls health/metrics, detects anomalies with explainable thresholds, and opens incidents only after consecutive failures.
- Used an LLM to separate root-cause services from downstream symptoms, then auto-remediated via a fixed whitelist (restart / traffic_shift / rate_limit / monitor) with confidence weighting, retry caps, and safe fallbacks — including real gateway traffic shifts, not simulated ones.
- Shipped a live dashboard with Break It + autonomous chaos, MTTD/MTTR metrics, incident timelines and LLM postmortems, plus 130+ tests, secret-leak CI checks, and SSRF-by-construction / rate-limit hardening for a public demo.
TypeScript, Node.js, Next.js, WebSockets, Docker Compose, Render, Vercel, GitHub Actions
- Built a distributed in-memory key-value store from scratch with Redis-style CRC16 hash slots, `MOVED`/`ASK` redirects, leader–follower replication, AOF + snapshot durability, pub/sub, and opt-in Raft-lite failover.
- Shipped a live cluster dashboard (topology, node health, replication lag, event feed, in-browser console) plus CLI and multi-process chaos tests (`SIGKILL` recovery, Compose smoke) so sharding/replication/failover are observable, not just documented.
- Deployed a reduced public demo (2-node Render + Vercel dashboard) with write-key protection, cold-start wake handling, and CI (build/test/audit/CodeQL/Gitleaks) for end-to-end portfolio proof without pretending it’s a Redis replacement.
Coding Agent
Node.js, Express, Groq (Llama 3.3), Supabase (PostgreSQL + pgvector), Hugging Face Embeddings, pg-boss, Next.js
- Built an autonomous bug-fixing agent that reads GitHub issues, locates relevant files via LLM-driven search, and generates and validates fixes in a sandbox with automated retries.
- Implemented a self-improving loop with vector memory (pgvector + Hugging Face embeddings) to track strategies across runs, plus surgical patching that commits only changed lines.
- Validated fixes with generated tests and ESLint before opening a draft PR with a confidence score, with full API/token cost tracking on a live dashboard.
Distributed API Engine
Node.js, Redis, MongoDB, Express, Next.js
- Developed a full-stack system to run API test collections asynchronously via a Redis + BullMQ queue-worker architecture with scalable job processing.
- Built deterministic job snapshotting, sequential workflow execution with variable chaining, and retry logic with exponential backoff.
- Secured the execution layer with SSRF protection, request validation, and encrypted credentials, backed by structured logging and metrics on a live Next.js dashboard.
RupeeRead
Python, FastAPI, Next.js, Supabase, Groq, Cohere, BM25
- Built an agentic RAG stack over Nifty 20 annual filings with hybrid retrieval (BM25 + pgvector + RRF), HyDE query expansion, multi-hop decomposition, LLM self-correction, and Cohere reranking, served via a FastAPI orchestrator streaming agent steps with page-level citations.
- Shipped an 11-section equity report agent with deterministic chart extraction, multi-company compare, and a faithfulness/hallucination eval pipeline, deployed end-to-end on Vercel (Next.js + API proxy) and Hugging Face Spaces (Docker) with rate limiting and CORS-locked security.
Rate Limiting Infrastructure
Node.js, Redis, Lua, Express, Next.js
- Built a Redis-backed distributed rate limiter using Lua scripts for atomic, race-free operations, supporting fixed window, sliding window, and token bucket algorithms.
- Implemented per-user and IP-based fallback rate limiting with configurable fail-open/fail-closed strategies for Redis outages.
- Developed a real-time monitoring dashboard with Next.js and Server-Sent Events, backed by automated tests and CI to validate rate limiting edge cases.
Workflow Agent
Node.js, Express, PostgreSQL (Neon), Prisma, pg-boss, Groq (Llama 3.3 70B), Slack API, GitHub API, Next.js
- Built a multi-agent orchestration system where an LLM classifier analyses GitHub/Slack webhook events and delegates to four specialist agents (data, analysis, comms, docs), each with a single-responsibility prompt.
- Implemented a durable PostgreSQL-backed job queue (pg-boss, no Redis) with automatic retries, an immutable audit trail, and full LLM cost/latency observability.
- Delivered end-to-end automation: GitHub issue/PR triage posts severity-coded Slack alerts and inline PR/issue comments, with a real-time Next.js dashboard streaming updates via Server-Sent Events.
Education
B.E in Artificial Intelligence And Data Science
Fr. Conceicao Rodrigues College Of Engineering
CGPA 8.22 · 2021 – 2025