Akash Chauhan
Resume

Full-Stack Developer & AI Systems Builder

I build AI systems you can trust.

Turning probabilistic language models into deterministic, benchmarked software architectures.

Noise into signal.

AI is noisy and probabilistic. Good engineering makes it deterministic and trusted.

Three rules. Three systems that prove them. ↓

Selected work

Rule 01

Separate extraction from math. LLMs read the messy text. Plain code does the maths.

GreenLedger-AI

ESG / BRSR sustainability reports.

Claude extracts cited numbers; Python does all GHG Scope 1/2, water and POSH maths. Titan Embeddings v2 + FAISS Top-5 cuts LLM input tokens ~40%. RAGAS: 1.0 context precision, 1.0 context recall, 0.86 faithfulness.

1.0
Context precision & recall
0.86
RAGAS faithfulness
~40%
Token reduction

React · Node · Python · FastAPI · AWS Bedrock · FAISS · BullMQ · S3

GitHub ↗

Rule 02

Evaluation-driven shipping. Every AI flow ships with a benchmark.

ClosedLoop

Autonomous self-healing knowledge base.

7-queue pipeline: ingestion + Luhn-validated PII redaction, semantic clustering, grounded drafting, holdout verification (temp-0 judge, 20% holdout, up to 3 rounds), atomic publish, live RAG deflection (0.78 cosine) over Socket.IO, knowledge-gap detection that spawns tickets.

7
Queue pipeline
46
Passing tests
0 / 784
PII leaks / records

React · Node · Express · MongoDB Atlas · Redis · BullMQ · Socket.IO · OpenAI

GitHub ↗

Rule 03

Product-first systems. Queues and indexes become fast, clear, trusted UI.

Ayu TokPress

Multi-provider AI chat for VS Code.

5 providers (Claude, OpenAI, Gemini, Copilot, Ollama). Streaming SSE, token-budget compression, keychain key storage. Live on VS Code Marketplace + Open VSX.

3k+
Downloads · Open VSX
5
AI providers
0 B
Ollama egress

TypeScript · VS Code API · esbuild · Ollama

VS Code Marketplace ↗ Open VSX ↗ Private repo

More work

Also built.

  • 04

    Placement Pilot

    Multi-agent placement intelligence.

    Custom ReAct agents; Recon queries FAISS before LLM reasoning; Strategy synthesizes preparation paths; Sentinel streams live risk alerts to the dashboard.

    3 ReAct agents · FAISS pre-LLM vector recall · Live Socket.IO risk stream

    Node · React · Redis · BullMQ · Socket.IO · Ollama · MongoDB · FAISS

    GitHub ↗

    Diagram: FAISS recall feeds the Recon agent, which hands off to Strategy and then Sentinel, which streams live risk alerts to the dashboard.
  • 05

    WhatsApp Credit Agent

    Autonomous invoice follow-up & classification (Sep – Oct 2026).

    Writes polite → firm → escalation follow-ups for overdue invoices, owner approves with one tap. LLM reply classifier: promise-to-pay date, dispute, silence.

    3-stage polite → firm → escalate · 1-tap owner approval · 3-class reply intent classifier

    Node · MongoDB · WhatsApp Business API · OpenAI · Gemini · Claude

    GitHub ↗

    Diagram: a three-stage follow-up ladder (polite, firm, escalate) with one-tap owner approval, then a reply classifier that sorts replies into promise-to-pay date, dispute and silence.
  • Exp 01

    Project Karlos

    Systematic trading testbed — deterministic execution rules, backtested signal pipelines, strict drawdown guards.

    Diagram: an equity curve whose losses are clipped at a drawdown guard line; trading halts when the curve crosses it.
  • Exp 02

    Trev

    Multi-modal mobility concept — unified routing and real-time transit handoff architecture.

    Building

    Diagram: a dashed multi-modal route network where a bus route and a rail route hand off at a central transit node. Status: building.

Research

LanguageFair.

Tokenization disparity across Indian languages for identical semantic payloads.

Hover a tower to separate its tokens. Drag to turn. Tap a tower to separate its tokens. Drag to turn.

Proof

Measured proof.

400+

DSA problems solved

1,654

Contest rating

Top 10

of 5,600+ teams — Cognizant Technoverse 2026

#2,480

of 47,935 — GSSoC 2026 rank

46

Tests passing · 0 PII leaks in 784 records

Experience

Software Developer Intern — Uplyx Solution

May – Jul 2025

  • Built RAG pipeline (LangChain + AWS Bedrock) behind Node.js REST APIs.
  • Custom ReAct loops, structured outputs, tool calling — 15% lower latency.
  • Evaluated 50+ queries with RAGAS; enforced hallucination and bias checks with guardrails.
  • Automated deployments with Jenkins + Docker CI/CD.

Certificates & honors

  • AWS Fundamentals of Generative AIOct 2026
  • Cognizant Technoverse 2026 Top 10 Finalist2026
  • AWS Cloud Practitioner EssentialsAug 2025
  • Oracle OCI Foundations AssociateAug 2025

Open source

GSSoC 2026 Contributor, Project Admin & Ambassador

Fixed XSS, DoS and prompt-injection issues in StudyPlan.

Education

Lovely Professional University

B.Tech in Computer Science & Engineering (2023 – 2027)

Contact

Let's build something that works.

Resume PDF GitHub ↗ LinkedIn ↗