anbukumaran/ tnea-advisor:latest

8 hours ago

TNEA-Advisor is a hyper-localized, data-driven admissions strategist mathematically fused into Qwen 2.5 3B and quantized to 4-bit GGUF for consumer hardware. It bypasses cloud dependencies to deliver hyper-accurate, offline engineering cutoff predictions. 

ollama run anbukumaran/tnea-advisor

Details

8 hours ago

5a105201d20f · 1.9GB ·

model
qwen2
·
3.09B
·
Q4_K_M
template
{{ if .System }}<|im_start|>system {{ .System }}<|im_end|> {{ end }}{{ if .Prompt }}<|im_start|>user
system
You are an expert TNEA engineering admissions counselor. You MUST communicate strictly in English. N
params
{ "stop": [ "<|im_end|>", "<|im_start|>" ], "temperature": 0.2, "top

Readme

🎓 TNEA-Advisor — Your Pocket TNEA Counselor

Runs entirely on your own laptop. No sign-up, no API key, no data ever leaving your machine.

⚡ Try it in one line

ollama run anbukumaran/tnea-advisor

A 3B-parameter model, quantized down to ~1.9 GB, that downloads once and then runs fully offline — no GPU required.

💬 Copy-paste this prompt

Cutoff: 198.50/200, Community: OC, Preferred Branch: CSE

Swap in your own cutoff mark, community category (OC / BC / BCM / MBC / SC / SCA / ST), and preferred branch — TNEA-Advisor takes it from there.

🔮 What you get back

TNEA-Advisor predicts your State General Rank and Community Rank, then sorts realistic colleges into three tiers so you instantly know where you stand:

Tier What it means
🚀 REACH Ambitious picks — your rank is right at or just above last year’s cutoff
🎯 TARGET Realistic bets — your rank comfortably clears the historical cutoff
🛡️ SAFE Confident backups — your rank sits well inside the cutoff range

Top-tier institutions like CEG, MIT, CIT Chennai, PSG, and SSN get flagged specifically when they’re in range.

🧠 Why it’s actually good at this

Most general chatbots will happily guess a rank and confidently invent a college list. TNEA-Advisor was fine-tuned specifically on real, category-wise historical DoTE cutoff data, then benchmarked against the un-tuned base model with an automated judge across 50 real scenarios:

  • ✅ 10 / 10 Accuracy — up 42.9% over the base model
  • ✅ 10 / 10 Hallucination Resistance — up 42.9%; it doesn’t invent ranks or colleges
  • ✅ 10 / 10 Relevance — up 25.0%

🪶 Why it’s so light

The model ships as a 4-bit quantized GGUF (q4_K_M) — about 1.9 GB — so it runs comfortably on a standard laptop. Everything happens locally: your marks and category never leave your device.

🛠️ Get the most accurate results

  • Give your exact cutoff mark — not a rounded estimate
  • Always state your community category — cutoffs vary a lot by category
  • Ask about one branch at a time for the sharpest tier predictions

🔗 More

Built and shipped by a TNEA aspirant — right before starting a BTech at CIT Chennai.